bnx: Defer interrupt allocation until chip reset is done
[dragonfly.git] / sys / dev / netif / bnx / if_bnx.c
1 /*
2  * Copyright (c) 2001 Wind River Systems
3  * Copyright (c) 1997, 1998, 1999, 2001
4  *      Bill Paul <wpaul@windriver.com>.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *      This product includes software developed by Bill Paul.
17  * 4. Neither the name of the author nor the names of any co-contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
25  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
31  * THE POSSIBILITY OF SUCH DAMAGE.
32  *
33  * $FreeBSD: src/sys/dev/bge/if_bge.c,v 1.3.2.39 2005/07/03 03:41:18 silby Exp $
34  */
35
36 #include "opt_bnx.h"
37 #include "opt_ifpoll.h"
38
39 #include <sys/param.h>
40 #include <sys/bus.h>
41 #include <sys/endian.h>
42 #include <sys/kernel.h>
43 #include <sys/interrupt.h>
44 #include <sys/mbuf.h>
45 #include <sys/malloc.h>
46 #include <sys/queue.h>
47 #include <sys/rman.h>
48 #include <sys/serialize.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/sysctl.h>
52
53 #include <netinet/ip.h>
54 #include <netinet/tcp.h>
55
56 #include <net/bpf.h>
57 #include <net/ethernet.h>
58 #include <net/if.h>
59 #include <net/if_arp.h>
60 #include <net/if_dl.h>
61 #include <net/if_media.h>
62 #include <net/if_poll.h>
63 #include <net/if_types.h>
64 #include <net/ifq_var.h>
65 #include <net/vlan/if_vlan_var.h>
66 #include <net/vlan/if_vlan_ether.h>
67
68 #include <dev/netif/mii_layer/mii.h>
69 #include <dev/netif/mii_layer/miivar.h>
70 #include <dev/netif/mii_layer/brgphyreg.h>
71
72 #include <bus/pci/pcidevs.h>
73 #include <bus/pci/pcireg.h>
74 #include <bus/pci/pcivar.h>
75
76 #include <dev/netif/bge/if_bgereg.h>
77 #include <dev/netif/bnx/if_bnxvar.h>
78
79 /* "device miibus" required.  See GENERIC if you get errors here. */
80 #include "miibus_if.h"
81
82 #define BNX_CSUM_FEATURES       (CSUM_IP | CSUM_TCP | CSUM_UDP)
83
84 #define BNX_INTR_CKINTVL        ((10 * hz) / 1000)      /* 10ms */
85
86 static const struct bnx_type {
87         uint16_t                bnx_vid;
88         uint16_t                bnx_did;
89         char                    *bnx_name;
90 } bnx_devs[] = {
91         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5717,
92                 "Broadcom BCM5717 Gigabit Ethernet" },
93         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5717C,
94                 "Broadcom BCM5717C Gigabit Ethernet" },
95         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5718,
96                 "Broadcom BCM5718 Gigabit Ethernet" },
97         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5719,
98                 "Broadcom BCM5719 Gigabit Ethernet" },
99         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5720_ALT,
100                 "Broadcom BCM5720 Gigabit Ethernet" },
101
102         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5725,
103                 "Broadcom BCM5725 Gigabit Ethernet" },
104         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5727,
105                 "Broadcom BCM5727 Gigabit Ethernet" },
106         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5762,
107                 "Broadcom BCM5762 Gigabit Ethernet" },
108
109         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57761,
110                 "Broadcom BCM57761 Gigabit Ethernet" },
111         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57762,
112                 "Broadcom BCM57762 Gigabit Ethernet" },
113         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57765,
114                 "Broadcom BCM57765 Gigabit Ethernet" },
115         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57766,
116                 "Broadcom BCM57766 Gigabit Ethernet" },
117         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57781,
118                 "Broadcom BCM57781 Gigabit Ethernet" },
119         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57782,
120                 "Broadcom BCM57782 Gigabit Ethernet" },
121         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57785,
122                 "Broadcom BCM57785 Gigabit Ethernet" },
123         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57786,
124                 "Broadcom BCM57786 Gigabit Ethernet" },
125         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57791,
126                 "Broadcom BCM57791 Fast Ethernet" },
127         { PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57795,
128                 "Broadcom BCM57795 Fast Ethernet" },
129
130         { 0, 0, NULL }
131 };
132
133 #define BNX_IS_JUMBO_CAPABLE(sc)        ((sc)->bnx_flags & BNX_FLAG_JUMBO)
134 #define BNX_IS_5717_PLUS(sc)            ((sc)->bnx_flags & BNX_FLAG_5717_PLUS)
135 #define BNX_IS_57765_PLUS(sc)           ((sc)->bnx_flags & BNX_FLAG_57765_PLUS)
136 #define BNX_IS_57765_FAMILY(sc)  \
137         ((sc)->bnx_flags & BNX_FLAG_57765_FAMILY)
138
139 typedef int     (*bnx_eaddr_fcn_t)(struct bnx_softc *, uint8_t[]);
140
141 static int      bnx_probe(device_t);
142 static int      bnx_attach(device_t);
143 static int      bnx_detach(device_t);
144 static void     bnx_shutdown(device_t);
145 static int      bnx_suspend(device_t);
146 static int      bnx_resume(device_t);
147 static int      bnx_miibus_readreg(device_t, int, int);
148 static int      bnx_miibus_writereg(device_t, int, int, int);
149 static void     bnx_miibus_statchg(device_t);
150
151 #ifdef IFPOLL_ENABLE
152 static void     bnx_npoll(struct ifnet *, struct ifpoll_info *);
153 static void     bnx_npoll_compat(struct ifnet *, void *, int);
154 #endif
155 static void     bnx_intr_legacy(void *);
156 static void     bnx_msi(void *);
157 static void     bnx_msi_oneshot(void *);
158 static void     bnx_intr(struct bnx_softc *);
159 static void     bnx_enable_intr(struct bnx_softc *);
160 static void     bnx_disable_intr(struct bnx_softc *);
161 static void     bnx_txeof(struct bnx_softc *, uint16_t);
162 static void     bnx_rxeof(struct bnx_softc *, uint16_t, int);
163
164 static void     bnx_start(struct ifnet *, struct ifaltq_subque *);
165 static int      bnx_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
166 static void     bnx_init(void *);
167 static void     bnx_stop(struct bnx_softc *);
168 static void     bnx_watchdog(struct ifnet *);
169 static int      bnx_ifmedia_upd(struct ifnet *);
170 static void     bnx_ifmedia_sts(struct ifnet *, struct ifmediareq *);
171 static void     bnx_tick(void *);
172
173 static int      bnx_alloc_jumbo_mem(struct bnx_softc *);
174 static void     bnx_free_jumbo_mem(struct bnx_softc *);
175 static struct bnx_jslot
176                 *bnx_jalloc(struct bnx_softc *);
177 static void     bnx_jfree(void *);
178 static void     bnx_jref(void *);
179 static int      bnx_newbuf_std(struct bnx_softc *, int, int);
180 static int      bnx_newbuf_jumbo(struct bnx_softc *, int, int);
181 static void     bnx_setup_rxdesc_std(struct bnx_softc *, int);
182 static void     bnx_setup_rxdesc_jumbo(struct bnx_softc *, int);
183 static int      bnx_init_rx_ring_std(struct bnx_softc *);
184 static void     bnx_free_rx_ring_std(struct bnx_softc *);
185 static int      bnx_init_rx_ring_jumbo(struct bnx_softc *);
186 static void     bnx_free_rx_ring_jumbo(struct bnx_softc *);
187 static void     bnx_free_tx_ring(struct bnx_softc *);
188 static int      bnx_init_tx_ring(struct bnx_softc *);
189 static int      bnx_dma_alloc(struct bnx_softc *);
190 static void     bnx_dma_free(struct bnx_softc *);
191 static int      bnx_dma_block_alloc(struct bnx_softc *, bus_size_t,
192                     bus_dma_tag_t *, bus_dmamap_t *, void **, bus_addr_t *);
193 static void     bnx_dma_block_free(bus_dma_tag_t, bus_dmamap_t, void *);
194 static struct mbuf *
195                 bnx_defrag_shortdma(struct mbuf *);
196 static int      bnx_encap(struct bnx_softc *, struct mbuf **,
197                         uint32_t *, int *);
198 static int      bnx_setup_tso(struct bnx_softc *, struct mbuf **,
199                     uint16_t *, uint16_t *);
200
201 static void     bnx_reset(struct bnx_softc *);
202 static int      bnx_chipinit(struct bnx_softc *);
203 static int      bnx_blockinit(struct bnx_softc *);
204 static void     bnx_stop_block(struct bnx_softc *, bus_size_t, uint32_t);
205 static void     bnx_enable_msi(struct bnx_softc *sc);
206 static void     bnx_setmulti(struct bnx_softc *);
207 static void     bnx_setpromisc(struct bnx_softc *);
208 static void     bnx_stats_update_regs(struct bnx_softc *);
209 static uint32_t bnx_dma_swap_options(struct bnx_softc *);
210
211 static uint32_t bnx_readmem_ind(struct bnx_softc *, uint32_t);
212 static void     bnx_writemem_ind(struct bnx_softc *, uint32_t, uint32_t);
213 #ifdef notdef
214 static uint32_t bnx_readreg_ind(struct bnx_softc *, uint32_t);
215 #endif
216 static void     bnx_writemem_direct(struct bnx_softc *, uint32_t, uint32_t);
217 static void     bnx_writembx(struct bnx_softc *, int, int);
218 static int      bnx_read_nvram(struct bnx_softc *, caddr_t, int, int);
219 static uint8_t  bnx_eeprom_getbyte(struct bnx_softc *, uint32_t, uint8_t *);
220 static int      bnx_read_eeprom(struct bnx_softc *, caddr_t, uint32_t, size_t);
221
222 static void     bnx_tbi_link_upd(struct bnx_softc *, uint32_t);
223 static void     bnx_copper_link_upd(struct bnx_softc *, uint32_t);
224 static void     bnx_autopoll_link_upd(struct bnx_softc *, uint32_t);
225 static void     bnx_link_poll(struct bnx_softc *);
226
227 static int      bnx_get_eaddr_mem(struct bnx_softc *, uint8_t[]);
228 static int      bnx_get_eaddr_nvram(struct bnx_softc *, uint8_t[]);
229 static int      bnx_get_eaddr_eeprom(struct bnx_softc *, uint8_t[]);
230 static int      bnx_get_eaddr(struct bnx_softc *, uint8_t[]);
231
232 static void     bnx_coal_change(struct bnx_softc *);
233 static int      bnx_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS);
234 static int      bnx_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS);
235 static int      bnx_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS);
236 static int      bnx_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS);
237 static int      bnx_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS);
238 static int      bnx_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS);
239 static int      bnx_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *,
240                     int, int, uint32_t);
241
242 static int      bnx_msi_enable = 1;
243 TUNABLE_INT("hw.bnx.msi.enable", &bnx_msi_enable);
244
245 static device_method_t bnx_methods[] = {
246         /* Device interface */
247         DEVMETHOD(device_probe,         bnx_probe),
248         DEVMETHOD(device_attach,        bnx_attach),
249         DEVMETHOD(device_detach,        bnx_detach),
250         DEVMETHOD(device_shutdown,      bnx_shutdown),
251         DEVMETHOD(device_suspend,       bnx_suspend),
252         DEVMETHOD(device_resume,        bnx_resume),
253
254         /* bus interface */
255         DEVMETHOD(bus_print_child,      bus_generic_print_child),
256         DEVMETHOD(bus_driver_added,     bus_generic_driver_added),
257
258         /* MII interface */
259         DEVMETHOD(miibus_readreg,       bnx_miibus_readreg),
260         DEVMETHOD(miibus_writereg,      bnx_miibus_writereg),
261         DEVMETHOD(miibus_statchg,       bnx_miibus_statchg),
262
263         DEVMETHOD_END
264 };
265
266 static DEFINE_CLASS_0(bnx, bnx_driver, bnx_methods, sizeof(struct bnx_softc));
267 static devclass_t bnx_devclass;
268
269 DECLARE_DUMMY_MODULE(if_bnx);
270 DRIVER_MODULE(if_bnx, pci, bnx_driver, bnx_devclass, NULL, NULL);
271 DRIVER_MODULE(miibus, bnx, miibus_driver, miibus_devclass, NULL, NULL);
272
273 static uint32_t
274 bnx_readmem_ind(struct bnx_softc *sc, uint32_t off)
275 {
276         device_t dev = sc->bnx_dev;
277         uint32_t val;
278
279         pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
280         val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4);
281         pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
282         return (val);
283 }
284
285 static void
286 bnx_writemem_ind(struct bnx_softc *sc, uint32_t off, uint32_t val)
287 {
288         device_t dev = sc->bnx_dev;
289
290         pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
291         pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4);
292         pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
293 }
294
295 static void
296 bnx_writemem_direct(struct bnx_softc *sc, uint32_t off, uint32_t val)
297 {
298         CSR_WRITE_4(sc, off, val);
299 }
300
301 static void
302 bnx_writembx(struct bnx_softc *sc, int off, int val)
303 {
304         CSR_WRITE_4(sc, off, val);
305 }
306
307 /*
308  * Read a sequence of bytes from NVRAM.
309  */
310 static int
311 bnx_read_nvram(struct bnx_softc *sc, caddr_t dest, int off, int cnt)
312 {
313         return (1);
314 }
315
316 /*
317  * Read a byte of data stored in the EEPROM at address 'addr.' The
318  * BCM570x supports both the traditional bitbang interface and an
319  * auto access interface for reading the EEPROM. We use the auto
320  * access method.
321  */
322 static uint8_t
323 bnx_eeprom_getbyte(struct bnx_softc *sc, uint32_t addr, uint8_t *dest)
324 {
325         int i;
326         uint32_t byte = 0;
327
328         /*
329          * Enable use of auto EEPROM access so we can avoid
330          * having to use the bitbang method.
331          */
332         BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
333
334         /* Reset the EEPROM, load the clock period. */
335         CSR_WRITE_4(sc, BGE_EE_ADDR,
336             BGE_EEADDR_RESET|BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
337         DELAY(20);
338
339         /* Issue the read EEPROM command. */
340         CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
341
342         /* Wait for completion */
343         for(i = 0; i < BNX_TIMEOUT * 10; i++) {
344                 DELAY(10);
345                 if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
346                         break;
347         }
348
349         if (i == BNX_TIMEOUT) {
350                 if_printf(&sc->arpcom.ac_if, "eeprom read timed out\n");
351                 return(1);
352         }
353
354         /* Get result. */
355         byte = CSR_READ_4(sc, BGE_EE_DATA);
356
357         *dest = (byte >> ((addr % 4) * 8)) & 0xFF;
358
359         return(0);
360 }
361
362 /*
363  * Read a sequence of bytes from the EEPROM.
364  */
365 static int
366 bnx_read_eeprom(struct bnx_softc *sc, caddr_t dest, uint32_t off, size_t len)
367 {
368         size_t i;
369         int err;
370         uint8_t byte;
371
372         for (byte = 0, err = 0, i = 0; i < len; i++) {
373                 err = bnx_eeprom_getbyte(sc, off + i, &byte);
374                 if (err)
375                         break;
376                 *(dest + i) = byte;
377         }
378
379         return(err ? 1 : 0);
380 }
381
382 static int
383 bnx_miibus_readreg(device_t dev, int phy, int reg)
384 {
385         struct bnx_softc *sc = device_get_softc(dev);
386         uint32_t val;
387         int i;
388
389         KASSERT(phy == sc->bnx_phyno,
390             ("invalid phyno %d, should be %d", phy, sc->bnx_phyno));
391
392         /* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
393         if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
394                 CSR_WRITE_4(sc, BGE_MI_MODE,
395                     sc->bnx_mi_mode & ~BGE_MIMODE_AUTOPOLL);
396                 DELAY(80);
397         }
398
399         CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
400             BGE_MIPHY(phy) | BGE_MIREG(reg));
401
402         /* Poll for the PHY register access to complete. */
403         for (i = 0; i < BNX_TIMEOUT; i++) {
404                 DELAY(10);
405                 val = CSR_READ_4(sc, BGE_MI_COMM);
406                 if ((val & BGE_MICOMM_BUSY) == 0) {
407                         DELAY(5);
408                         val = CSR_READ_4(sc, BGE_MI_COMM);
409                         break;
410                 }
411         }
412         if (i == BNX_TIMEOUT) {
413                 if_printf(&sc->arpcom.ac_if, "PHY read timed out "
414                     "(phy %d, reg %d, val 0x%08x)\n", phy, reg, val);
415                 val = 0;
416         }
417
418         /* Restore the autopoll bit if necessary. */
419         if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
420                 CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
421                 DELAY(80);
422         }
423
424         if (val & BGE_MICOMM_READFAIL)
425                 return 0;
426
427         return (val & 0xFFFF);
428 }
429
430 static int
431 bnx_miibus_writereg(device_t dev, int phy, int reg, int val)
432 {
433         struct bnx_softc *sc = device_get_softc(dev);
434         int i;
435
436         KASSERT(phy == sc->bnx_phyno,
437             ("invalid phyno %d, should be %d", phy, sc->bnx_phyno));
438
439         /* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
440         if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
441                 CSR_WRITE_4(sc, BGE_MI_MODE,
442                     sc->bnx_mi_mode & ~BGE_MIMODE_AUTOPOLL);
443                 DELAY(80);
444         }
445
446         CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
447             BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
448
449         for (i = 0; i < BNX_TIMEOUT; i++) {
450                 DELAY(10);
451                 if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
452                         DELAY(5);
453                         CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */
454                         break;
455                 }
456         }
457         if (i == BNX_TIMEOUT) {
458                 if_printf(&sc->arpcom.ac_if, "PHY write timed out "
459                     "(phy %d, reg %d, val %d)\n", phy, reg, val);
460         }
461
462         /* Restore the autopoll bit if necessary. */
463         if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
464                 CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
465                 DELAY(80);
466         }
467
468         return 0;
469 }
470
471 static void
472 bnx_miibus_statchg(device_t dev)
473 {
474         struct bnx_softc *sc;
475         struct mii_data *mii;
476
477         sc = device_get_softc(dev);
478         mii = device_get_softc(sc->bnx_miibus);
479
480         if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
481             (IFM_ACTIVE | IFM_AVALID)) {
482                 switch (IFM_SUBTYPE(mii->mii_media_active)) {
483                 case IFM_10_T:
484                 case IFM_100_TX:
485                         sc->bnx_link = 1;
486                         break;
487                 case IFM_1000_T:
488                 case IFM_1000_SX:
489                 case IFM_2500_SX:
490                         sc->bnx_link = 1;
491                         break;
492                 default:
493                         sc->bnx_link = 0;
494                         break;
495                 }
496         } else {
497                 sc->bnx_link = 0;
498         }
499         if (sc->bnx_link == 0)
500                 return;
501
502         BNX_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
503         if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
504             IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX) {
505                 BNX_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
506         } else {
507                 BNX_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
508         }
509
510         if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX) {
511                 BNX_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
512         } else {
513                 BNX_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
514         }
515 }
516
517 /*
518  * Memory management for jumbo frames.
519  */
520 static int
521 bnx_alloc_jumbo_mem(struct bnx_softc *sc)
522 {
523         struct ifnet *ifp = &sc->arpcom.ac_if;
524         struct bnx_jslot *entry;
525         uint8_t *ptr;
526         bus_addr_t paddr;
527         int i, error;
528
529         /*
530          * Create tag for jumbo mbufs.
531          * This is really a bit of a kludge. We allocate a special
532          * jumbo buffer pool which (thanks to the way our DMA
533          * memory allocation works) will consist of contiguous
534          * pages. This means that even though a jumbo buffer might
535          * be larger than a page size, we don't really need to
536          * map it into more than one DMA segment. However, the
537          * default mbuf tag will result in multi-segment mappings,
538          * so we have to create a special jumbo mbuf tag that
539          * lets us get away with mapping the jumbo buffers as
540          * a single segment. I think eventually the driver should
541          * be changed so that it uses ordinary mbufs and cluster
542          * buffers, i.e. jumbo frames can span multiple DMA
543          * descriptors. But that's a project for another day.
544          */
545
546         /*
547          * Create DMA stuffs for jumbo RX ring.
548          */
549         error = bnx_dma_block_alloc(sc, BGE_JUMBO_RX_RING_SZ,
550                                     &sc->bnx_cdata.bnx_rx_jumbo_ring_tag,
551                                     &sc->bnx_cdata.bnx_rx_jumbo_ring_map,
552                                     (void *)&sc->bnx_ldata.bnx_rx_jumbo_ring,
553                                     &sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
554         if (error) {
555                 if_printf(ifp, "could not create jumbo RX ring\n");
556                 return error;
557         }
558
559         /*
560          * Create DMA stuffs for jumbo buffer block.
561          */
562         error = bnx_dma_block_alloc(sc, BNX_JMEM,
563                                     &sc->bnx_cdata.bnx_jumbo_tag,
564                                     &sc->bnx_cdata.bnx_jumbo_map,
565                                     (void **)&sc->bnx_ldata.bnx_jumbo_buf,
566                                     &paddr);
567         if (error) {
568                 if_printf(ifp, "could not create jumbo buffer\n");
569                 return error;
570         }
571
572         SLIST_INIT(&sc->bnx_jfree_listhead);
573
574         /*
575          * Now divide it up into 9K pieces and save the addresses
576          * in an array. Note that we play an evil trick here by using
577          * the first few bytes in the buffer to hold the the address
578          * of the softc structure for this interface. This is because
579          * bnx_jfree() needs it, but it is called by the mbuf management
580          * code which will not pass it to us explicitly.
581          */
582         for (i = 0, ptr = sc->bnx_ldata.bnx_jumbo_buf; i < BNX_JSLOTS; i++) {
583                 entry = &sc->bnx_cdata.bnx_jslots[i];
584                 entry->bnx_sc = sc;
585                 entry->bnx_buf = ptr;
586                 entry->bnx_paddr = paddr;
587                 entry->bnx_inuse = 0;
588                 entry->bnx_slot = i;
589                 SLIST_INSERT_HEAD(&sc->bnx_jfree_listhead, entry, jslot_link);
590
591                 ptr += BNX_JLEN;
592                 paddr += BNX_JLEN;
593         }
594         return 0;
595 }
596
597 static void
598 bnx_free_jumbo_mem(struct bnx_softc *sc)
599 {
600         /* Destroy jumbo RX ring. */
601         bnx_dma_block_free(sc->bnx_cdata.bnx_rx_jumbo_ring_tag,
602                            sc->bnx_cdata.bnx_rx_jumbo_ring_map,
603                            sc->bnx_ldata.bnx_rx_jumbo_ring);
604
605         /* Destroy jumbo buffer block. */
606         bnx_dma_block_free(sc->bnx_cdata.bnx_jumbo_tag,
607                            sc->bnx_cdata.bnx_jumbo_map,
608                            sc->bnx_ldata.bnx_jumbo_buf);
609 }
610
611 /*
612  * Allocate a jumbo buffer.
613  */
614 static struct bnx_jslot *
615 bnx_jalloc(struct bnx_softc *sc)
616 {
617         struct bnx_jslot *entry;
618
619         lwkt_serialize_enter(&sc->bnx_jslot_serializer);
620         entry = SLIST_FIRST(&sc->bnx_jfree_listhead);
621         if (entry) {
622                 SLIST_REMOVE_HEAD(&sc->bnx_jfree_listhead, jslot_link);
623                 entry->bnx_inuse = 1;
624         } else {
625                 if_printf(&sc->arpcom.ac_if, "no free jumbo buffers\n");
626         }
627         lwkt_serialize_exit(&sc->bnx_jslot_serializer);
628         return(entry);
629 }
630
631 /*
632  * Adjust usage count on a jumbo buffer.
633  */
634 static void
635 bnx_jref(void *arg)
636 {
637         struct bnx_jslot *entry = (struct bnx_jslot *)arg;
638         struct bnx_softc *sc = entry->bnx_sc;
639
640         if (sc == NULL)
641                 panic("bnx_jref: can't find softc pointer!");
642
643         if (&sc->bnx_cdata.bnx_jslots[entry->bnx_slot] != entry) {
644                 panic("bnx_jref: asked to reference buffer "
645                     "that we don't manage!");
646         } else if (entry->bnx_inuse == 0) {
647                 panic("bnx_jref: buffer already free!");
648         } else {
649                 atomic_add_int(&entry->bnx_inuse, 1);
650         }
651 }
652
653 /*
654  * Release a jumbo buffer.
655  */
656 static void
657 bnx_jfree(void *arg)
658 {
659         struct bnx_jslot *entry = (struct bnx_jslot *)arg;
660         struct bnx_softc *sc = entry->bnx_sc;
661
662         if (sc == NULL)
663                 panic("bnx_jfree: can't find softc pointer!");
664
665         if (&sc->bnx_cdata.bnx_jslots[entry->bnx_slot] != entry) {
666                 panic("bnx_jfree: asked to free buffer that we don't manage!");
667         } else if (entry->bnx_inuse == 0) {
668                 panic("bnx_jfree: buffer already free!");
669         } else {
670                 /*
671                  * Possible MP race to 0, use the serializer.  The atomic insn
672                  * is still needed for races against bnx_jref().
673                  */
674                 lwkt_serialize_enter(&sc->bnx_jslot_serializer);
675                 atomic_subtract_int(&entry->bnx_inuse, 1);
676                 if (entry->bnx_inuse == 0) {
677                         SLIST_INSERT_HEAD(&sc->bnx_jfree_listhead, 
678                                           entry, jslot_link);
679                 }
680                 lwkt_serialize_exit(&sc->bnx_jslot_serializer);
681         }
682 }
683
684
685 /*
686  * Intialize a standard receive ring descriptor.
687  */
688 static int
689 bnx_newbuf_std(struct bnx_softc *sc, int i, int init)
690 {
691         struct mbuf *m_new = NULL;
692         bus_dma_segment_t seg;
693         bus_dmamap_t map;
694         int error, nsegs;
695
696         m_new = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
697         if (m_new == NULL)
698                 return ENOBUFS;
699         m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
700         m_adj(m_new, ETHER_ALIGN);
701
702         error = bus_dmamap_load_mbuf_segment(sc->bnx_cdata.bnx_rx_mtag,
703                         sc->bnx_cdata.bnx_rx_tmpmap, m_new,
704                         &seg, 1, &nsegs, BUS_DMA_NOWAIT);
705         if (error) {
706                 m_freem(m_new);
707                 return error;
708         }
709
710         if (!init) {
711                 bus_dmamap_sync(sc->bnx_cdata.bnx_rx_mtag,
712                                 sc->bnx_cdata.bnx_rx_std_dmamap[i],
713                                 BUS_DMASYNC_POSTREAD);
714                 bus_dmamap_unload(sc->bnx_cdata.bnx_rx_mtag,
715                         sc->bnx_cdata.bnx_rx_std_dmamap[i]);
716         }
717
718         map = sc->bnx_cdata.bnx_rx_tmpmap;
719         sc->bnx_cdata.bnx_rx_tmpmap = sc->bnx_cdata.bnx_rx_std_dmamap[i];
720         sc->bnx_cdata.bnx_rx_std_dmamap[i] = map;
721
722         sc->bnx_cdata.bnx_rx_std_chain[i].bnx_mbuf = m_new;
723         sc->bnx_cdata.bnx_rx_std_chain[i].bnx_paddr = seg.ds_addr;
724
725         bnx_setup_rxdesc_std(sc, i);
726         return 0;
727 }
728
729 static void
730 bnx_setup_rxdesc_std(struct bnx_softc *sc, int i)
731 {
732         struct bnx_rxchain *rc;
733         struct bge_rx_bd *r;
734
735         rc = &sc->bnx_cdata.bnx_rx_std_chain[i];
736         r = &sc->bnx_ldata.bnx_rx_std_ring[i];
737
738         r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bnx_paddr);
739         r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bnx_paddr);
740         r->bge_len = rc->bnx_mbuf->m_len;
741         r->bge_idx = i;
742         r->bge_flags = BGE_RXBDFLAG_END;
743 }
744
745 /*
746  * Initialize a jumbo receive ring descriptor. This allocates
747  * a jumbo buffer from the pool managed internally by the driver.
748  */
749 static int
750 bnx_newbuf_jumbo(struct bnx_softc *sc, int i, int init)
751 {
752         struct mbuf *m_new = NULL;
753         struct bnx_jslot *buf;
754         bus_addr_t paddr;
755
756         /* Allocate the mbuf. */
757         MGETHDR(m_new, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
758         if (m_new == NULL)
759                 return ENOBUFS;
760
761         /* Allocate the jumbo buffer */
762         buf = bnx_jalloc(sc);
763         if (buf == NULL) {
764                 m_freem(m_new);
765                 return ENOBUFS;
766         }
767
768         /* Attach the buffer to the mbuf. */
769         m_new->m_ext.ext_arg = buf;
770         m_new->m_ext.ext_buf = buf->bnx_buf;
771         m_new->m_ext.ext_free = bnx_jfree;
772         m_new->m_ext.ext_ref = bnx_jref;
773         m_new->m_ext.ext_size = BNX_JUMBO_FRAMELEN;
774
775         m_new->m_flags |= M_EXT;
776
777         m_new->m_data = m_new->m_ext.ext_buf;
778         m_new->m_len = m_new->m_pkthdr.len = m_new->m_ext.ext_size;
779
780         paddr = buf->bnx_paddr;
781         m_adj(m_new, ETHER_ALIGN);
782         paddr += ETHER_ALIGN;
783
784         /* Save necessary information */
785         sc->bnx_cdata.bnx_rx_jumbo_chain[i].bnx_mbuf = m_new;
786         sc->bnx_cdata.bnx_rx_jumbo_chain[i].bnx_paddr = paddr;
787
788         /* Set up the descriptor. */
789         bnx_setup_rxdesc_jumbo(sc, i);
790         return 0;
791 }
792
793 static void
794 bnx_setup_rxdesc_jumbo(struct bnx_softc *sc, int i)
795 {
796         struct bge_rx_bd *r;
797         struct bnx_rxchain *rc;
798
799         r = &sc->bnx_ldata.bnx_rx_jumbo_ring[i];
800         rc = &sc->bnx_cdata.bnx_rx_jumbo_chain[i];
801
802         r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bnx_paddr);
803         r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bnx_paddr);
804         r->bge_len = rc->bnx_mbuf->m_len;
805         r->bge_idx = i;
806         r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
807 }
808
809 static int
810 bnx_init_rx_ring_std(struct bnx_softc *sc)
811 {
812         int i, error;
813
814         for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
815                 error = bnx_newbuf_std(sc, i, 1);
816                 if (error)
817                         return error;
818         }
819
820         sc->bnx_std = BGE_STD_RX_RING_CNT - 1;
821         bnx_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bnx_std);
822
823         return(0);
824 }
825
826 static void
827 bnx_free_rx_ring_std(struct bnx_softc *sc)
828 {
829         int i;
830
831         for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
832                 struct bnx_rxchain *rc = &sc->bnx_cdata.bnx_rx_std_chain[i];
833
834                 if (rc->bnx_mbuf != NULL) {
835                         bus_dmamap_unload(sc->bnx_cdata.bnx_rx_mtag,
836                                           sc->bnx_cdata.bnx_rx_std_dmamap[i]);
837                         m_freem(rc->bnx_mbuf);
838                         rc->bnx_mbuf = NULL;
839                 }
840                 bzero(&sc->bnx_ldata.bnx_rx_std_ring[i],
841                     sizeof(struct bge_rx_bd));
842         }
843 }
844
845 static int
846 bnx_init_rx_ring_jumbo(struct bnx_softc *sc)
847 {
848         struct bge_rcb *rcb;
849         int i, error;
850
851         for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
852                 error = bnx_newbuf_jumbo(sc, i, 1);
853                 if (error)
854                         return error;
855         }
856
857         sc->bnx_jumbo = BGE_JUMBO_RX_RING_CNT - 1;
858
859         rcb = &sc->bnx_ldata.bnx_info.bnx_jumbo_rx_rcb;
860         rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, 0);
861         CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
862
863         bnx_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bnx_jumbo);
864
865         return(0);
866 }
867
868 static void
869 bnx_free_rx_ring_jumbo(struct bnx_softc *sc)
870 {
871         int i;
872
873         for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
874                 struct bnx_rxchain *rc = &sc->bnx_cdata.bnx_rx_jumbo_chain[i];
875
876                 if (rc->bnx_mbuf != NULL) {
877                         m_freem(rc->bnx_mbuf);
878                         rc->bnx_mbuf = NULL;
879                 }
880                 bzero(&sc->bnx_ldata.bnx_rx_jumbo_ring[i],
881                     sizeof(struct bge_rx_bd));
882         }
883 }
884
885 static void
886 bnx_free_tx_ring(struct bnx_softc *sc)
887 {
888         int i;
889
890         for (i = 0; i < BGE_TX_RING_CNT; i++) {
891                 if (sc->bnx_cdata.bnx_tx_chain[i] != NULL) {
892                         bus_dmamap_unload(sc->bnx_cdata.bnx_tx_mtag,
893                                           sc->bnx_cdata.bnx_tx_dmamap[i]);
894                         m_freem(sc->bnx_cdata.bnx_tx_chain[i]);
895                         sc->bnx_cdata.bnx_tx_chain[i] = NULL;
896                 }
897                 bzero(&sc->bnx_ldata.bnx_tx_ring[i],
898                     sizeof(struct bge_tx_bd));
899         }
900 }
901
902 static int
903 bnx_init_tx_ring(struct bnx_softc *sc)
904 {
905         sc->bnx_txcnt = 0;
906         sc->bnx_tx_saved_considx = 0;
907         sc->bnx_tx_prodidx = 0;
908
909         /* Initialize transmit producer index for host-memory send ring. */
910         bnx_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bnx_tx_prodidx);
911         bnx_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
912
913         return(0);
914 }
915
916 static void
917 bnx_setmulti(struct bnx_softc *sc)
918 {
919         struct ifnet *ifp;
920         struct ifmultiaddr *ifma;
921         uint32_t hashes[4] = { 0, 0, 0, 0 };
922         int h, i;
923
924         ifp = &sc->arpcom.ac_if;
925
926         if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
927                 for (i = 0; i < 4; i++)
928                         CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF);
929                 return;
930         }
931
932         /* First, zot all the existing filters. */
933         for (i = 0; i < 4; i++)
934                 CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0);
935
936         /* Now program new ones. */
937         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
938                 if (ifma->ifma_addr->sa_family != AF_LINK)
939                         continue;
940                 h = ether_crc32_le(
941                     LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
942                     ETHER_ADDR_LEN) & 0x7f;
943                 hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
944         }
945
946         for (i = 0; i < 4; i++)
947                 CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
948 }
949
950 /*
951  * Do endian, PCI and DMA initialization. Also check the on-board ROM
952  * self-test results.
953  */
954 static int
955 bnx_chipinit(struct bnx_softc *sc)
956 {
957         uint32_t dma_rw_ctl, mode_ctl;
958         int i;
959
960         /* Set endian type before we access any non-PCI registers. */
961         pci_write_config(sc->bnx_dev, BGE_PCI_MISC_CTL,
962             BGE_INIT | BGE_PCIMISCCTL_TAGGED_STATUS, 4);
963
964         /* Clear the MAC control register */
965         CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
966
967         /*
968          * Clear the MAC statistics block in the NIC's
969          * internal memory.
970          */
971         for (i = BGE_STATS_BLOCK;
972             i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
973                 BNX_MEMWIN_WRITE(sc, i, 0);
974
975         for (i = BGE_STATUS_BLOCK;
976             i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
977                 BNX_MEMWIN_WRITE(sc, i, 0);
978
979         if (BNX_IS_57765_FAMILY(sc)) {
980                 uint32_t val;
981
982                 if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0) {
983                         mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
984                         val = mode_ctl & ~BGE_MODECTL_PCIE_PORTS;
985
986                         /* Access the lower 1K of PL PCI-E block registers. */
987                         CSR_WRITE_4(sc, BGE_MODE_CTL,
988                             val | BGE_MODECTL_PCIE_PL_SEL);
989
990                         val = CSR_READ_4(sc, BGE_PCIE_PL_LO_PHYCTL5);
991                         val |= BGE_PCIE_PL_LO_PHYCTL5_DIS_L2CLKREQ;
992                         CSR_WRITE_4(sc, BGE_PCIE_PL_LO_PHYCTL5, val);
993
994                         CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
995                 }
996                 if (sc->bnx_chiprev != BGE_CHIPREV_57765_AX) {
997                         /* Fix transmit hangs */
998                         val = CSR_READ_4(sc, BGE_CPMU_PADRNG_CTL);
999                         val |= BGE_CPMU_PADRNG_CTL_RDIV2;
1000                         CSR_WRITE_4(sc, BGE_CPMU_PADRNG_CTL, val);
1001
1002                         mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
1003                         val = mode_ctl & ~BGE_MODECTL_PCIE_PORTS;
1004
1005                         /* Access the lower 1K of DL PCI-E block registers. */
1006                         CSR_WRITE_4(sc, BGE_MODE_CTL,
1007                             val | BGE_MODECTL_PCIE_DL_SEL);
1008
1009                         val = CSR_READ_4(sc, BGE_PCIE_DL_LO_FTSMAX);
1010                         val &= ~BGE_PCIE_DL_LO_FTSMAX_MASK;
1011                         val |= BGE_PCIE_DL_LO_FTSMAX_VAL;
1012                         CSR_WRITE_4(sc, BGE_PCIE_DL_LO_FTSMAX, val);
1013
1014                         CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1015                 }
1016
1017                 val = CSR_READ_4(sc, BGE_CPMU_LSPD_10MB_CLK);
1018                 val &= ~BGE_CPMU_LSPD_10MB_MACCLK_MASK;
1019                 val |= BGE_CPMU_LSPD_10MB_MACCLK_6_25;
1020                 CSR_WRITE_4(sc, BGE_CPMU_LSPD_10MB_CLK, val);
1021         }
1022
1023         /*
1024          * Set up the PCI DMA control register.
1025          */
1026         dma_rw_ctl = pci_read_config(sc->bnx_dev, BGE_PCI_DMA_RW_CTL, 4);
1027         /*
1028          * Disable 32bytes cache alignment for DMA write to host memory
1029          *
1030          * NOTE:
1031          * 64bytes cache alignment for DMA write to host memory is still
1032          * enabled.
1033          */
1034         dma_rw_ctl |= BGE_PCIDMARWCTL_DIS_CACHE_ALIGNMENT;
1035         if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0)
1036                 dma_rw_ctl &= ~BGE_PCIDMARWCTL_CRDRDR_RDMA_MRRS_MSK;
1037         /*
1038          * Enable HW workaround for controllers that misinterpret
1039          * a status tag update and leave interrupts permanently
1040          * disabled.
1041          */
1042         if (sc->bnx_asicrev != BGE_ASICREV_BCM5717 &&
1043             sc->bnx_asicrev != BGE_ASICREV_BCM5762 &&
1044             !BNX_IS_57765_FAMILY(sc))
1045                 dma_rw_ctl |= BGE_PCIDMARWCTL_TAGGED_STATUS_WA;
1046         if (bootverbose) {
1047                 if_printf(&sc->arpcom.ac_if, "DMA read/write %#x\n",
1048                     dma_rw_ctl);
1049         }
1050         pci_write_config(sc->bnx_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4);
1051
1052         /*
1053          * Set up general mode register.
1054          */
1055         mode_ctl = bnx_dma_swap_options(sc) | BGE_MODECTL_MAC_ATTN_INTR |
1056             BGE_MODECTL_HOST_SEND_BDS | BGE_MODECTL_TX_NO_PHDR_CSUM;
1057         CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1058
1059         /*
1060          * Disable memory write invalidate.  Apparently it is not supported
1061          * properly by these devices.  Also ensure that INTx isn't disabled,
1062          * as these chips need it even when using MSI.
1063          */
1064         PCI_CLRBIT(sc->bnx_dev, BGE_PCI_CMD,
1065             (PCIM_CMD_MWRICEN | PCIM_CMD_INTxDIS), 4);
1066
1067         /* Set the timer prescaler (always 66Mhz) */
1068         CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1069
1070         return(0);
1071 }
1072
1073 static int
1074 bnx_blockinit(struct bnx_softc *sc)
1075 {
1076         struct bge_rcb *rcb;
1077         bus_size_t vrcb;
1078         bge_hostaddr taddr;
1079         uint32_t val;
1080         int i, limit;
1081
1082         /*
1083          * Initialize the memory window pointer register so that
1084          * we can access the first 32K of internal NIC RAM. This will
1085          * allow us to set up the TX send ring RCBs and the RX return
1086          * ring RCBs, plus other things which live in NIC memory.
1087          */
1088         CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0);
1089
1090         /* Configure mbuf pool watermarks */
1091         if (BNX_IS_57765_PLUS(sc)) {
1092                 CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1093                 if (sc->arpcom.ac_if.if_mtu > ETHERMTU) {
1094                         CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x7e);
1095                         CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xea);
1096                 } else {
1097                         CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x2a);
1098                         CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xa0);
1099                 }
1100         } else {
1101                 CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1102                 CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1103                 CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1104         }
1105
1106         /* Configure DMA resource watermarks */
1107         CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1108         CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1109
1110         /* Enable buffer manager */
1111         val = BGE_BMANMODE_ENABLE | BGE_BMANMODE_LOMBUF_ATTN;
1112         /*
1113          * Change the arbitration algorithm of TXMBUF read request to
1114          * round-robin instead of priority based for BCM5719.  When
1115          * TXFIFO is almost empty, RDMA will hold its request until
1116          * TXFIFO is not almost empty.
1117          */
1118         if (sc->bnx_asicrev == BGE_ASICREV_BCM5719)
1119                 val |= BGE_BMANMODE_NO_TX_UNDERRUN;
1120         if (sc->bnx_asicrev == BGE_ASICREV_BCM5717 ||
1121             sc->bnx_chipid == BGE_CHIPID_BCM5719_A0 ||
1122             sc->bnx_chipid == BGE_CHIPID_BCM5720_A0)
1123                 val |= BGE_BMANMODE_LOMBUF_ATTN;
1124         CSR_WRITE_4(sc, BGE_BMAN_MODE, val);
1125
1126         /* Poll for buffer manager start indication */
1127         for (i = 0; i < BNX_TIMEOUT; i++) {
1128                 if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1129                         break;
1130                 DELAY(10);
1131         }
1132
1133         if (i == BNX_TIMEOUT) {
1134                 if_printf(&sc->arpcom.ac_if,
1135                           "buffer manager failed to start\n");
1136                 return(ENXIO);
1137         }
1138
1139         /* Enable flow-through queues */
1140         CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1141         CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1142
1143         /* Wait until queue initialization is complete */
1144         for (i = 0; i < BNX_TIMEOUT; i++) {
1145                 if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1146                         break;
1147                 DELAY(10);
1148         }
1149
1150         if (i == BNX_TIMEOUT) {
1151                 if_printf(&sc->arpcom.ac_if,
1152                           "flow-through queue init failed\n");
1153                 return(ENXIO);
1154         }
1155
1156         /*
1157          * Summary of rings supported by the controller:
1158          *
1159          * Standard Receive Producer Ring
1160          * - This ring is used to feed receive buffers for "standard"
1161          *   sized frames (typically 1536 bytes) to the controller.
1162          *
1163          * Jumbo Receive Producer Ring
1164          * - This ring is used to feed receive buffers for jumbo sized
1165          *   frames (i.e. anything bigger than the "standard" frames)
1166          *   to the controller.
1167          *
1168          * Mini Receive Producer Ring
1169          * - This ring is used to feed receive buffers for "mini"
1170          *   sized frames to the controller.
1171          * - This feature required external memory for the controller
1172          *   but was never used in a production system.  Should always
1173          *   be disabled.
1174          *
1175          * Receive Return Ring
1176          * - After the controller has placed an incoming frame into a
1177          *   receive buffer that buffer is moved into a receive return
1178          *   ring.  The driver is then responsible to passing the
1179          *   buffer up to the stack.  Many versions of the controller
1180          *   support multiple RR rings.
1181          *
1182          * Send Ring
1183          * - This ring is used for outgoing frames.  Many versions of
1184          *   the controller support multiple send rings.
1185          */
1186
1187         /* Initialize the standard receive producer ring control block. */
1188         rcb = &sc->bnx_ldata.bnx_info.bnx_std_rx_rcb;
1189         rcb->bge_hostaddr.bge_addr_lo =
1190             BGE_ADDR_LO(sc->bnx_ldata.bnx_rx_std_ring_paddr);
1191         rcb->bge_hostaddr.bge_addr_hi =
1192             BGE_ADDR_HI(sc->bnx_ldata.bnx_rx_std_ring_paddr);
1193         if (BNX_IS_57765_PLUS(sc)) {
1194                 /*
1195                  * Bits 31-16: Programmable ring size (2048, 1024, 512, .., 32)
1196                  * Bits 15-2 : Maximum RX frame size
1197                  * Bit 1     : 1 = Ring Disabled, 0 = Ring ENabled
1198                  * Bit 0     : Reserved
1199                  */
1200                 rcb->bge_maxlen_flags =
1201                     BGE_RCB_MAXLEN_FLAGS(512, BNX_MAX_FRAMELEN << 2);
1202         } else {
1203                 /*
1204                  * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
1205                  * Bits 15-2 : Reserved (should be 0)
1206                  * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1207                  * Bit 0     : Reserved
1208                  */
1209                 rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1210         }
1211         if (BNX_IS_5717_PLUS(sc))
1212                 rcb->bge_nicaddr = BGE_STD_RX_RINGS_5717;
1213         else
1214                 rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1215         /* Write the standard receive producer ring control block. */
1216         CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1217         CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1218         CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1219         CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1220         /* Reset the standard receive producer ring producer index. */
1221         bnx_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1222
1223         /*
1224          * Initialize the jumbo RX producer ring control
1225          * block.  We set the 'ring disabled' bit in the
1226          * flags field until we're actually ready to start
1227          * using this ring (i.e. once we set the MTU
1228          * high enough to require it).
1229          */
1230         if (BNX_IS_JUMBO_CAPABLE(sc)) {
1231                 rcb = &sc->bnx_ldata.bnx_info.bnx_jumbo_rx_rcb;
1232                 /* Get the jumbo receive producer ring RCB parameters. */
1233                 rcb->bge_hostaddr.bge_addr_lo =
1234                     BGE_ADDR_LO(sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
1235                 rcb->bge_hostaddr.bge_addr_hi =
1236                     BGE_ADDR_HI(sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
1237                 rcb->bge_maxlen_flags =
1238                     BGE_RCB_MAXLEN_FLAGS(BNX_MAX_FRAMELEN,
1239                     BGE_RCB_FLAG_RING_DISABLED);
1240                 if (BNX_IS_5717_PLUS(sc))
1241                         rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS_5717;
1242                 else
1243                         rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1244                 CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1245                     rcb->bge_hostaddr.bge_addr_hi);
1246                 CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1247                     rcb->bge_hostaddr.bge_addr_lo);
1248                 /* Program the jumbo receive producer ring RCB parameters. */
1249                 CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1250                     rcb->bge_maxlen_flags);
1251                 CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1252                 /* Reset the jumbo receive producer ring producer index. */
1253                 bnx_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1254         }
1255
1256         /*
1257          * The BD ring replenish thresholds control how often the
1258          * hardware fetches new BD's from the producer rings in host
1259          * memory.  Setting the value too low on a busy system can
1260          * starve the hardware and recue the throughpout.
1261          *
1262          * Set the BD ring replentish thresholds. The recommended
1263          * values are 1/8th the number of descriptors allocated to
1264          * each ring.
1265          */
1266         val = 8;
1267         CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val);
1268         if (BNX_IS_JUMBO_CAPABLE(sc)) {
1269                 CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH,
1270                     BGE_JUMBO_RX_RING_CNT/8);
1271         }
1272         if (BNX_IS_57765_PLUS(sc)) {
1273                 CSR_WRITE_4(sc, BGE_STD_REPLENISH_LWM, 32);
1274                 CSR_WRITE_4(sc, BGE_JMB_REPLENISH_LWM, 16);
1275         }
1276
1277         /*
1278          * Disable all send rings by setting the 'ring disabled' bit
1279          * in the flags field of all the TX send ring control blocks,
1280          * located in NIC memory.
1281          */
1282         if (BNX_IS_5717_PLUS(sc))
1283                 limit = 4;
1284         else if (BNX_IS_57765_FAMILY(sc) ||
1285             sc->bnx_asicrev == BGE_ASICREV_BCM5762)
1286                 limit = 2;
1287         else
1288                 limit = 1;
1289         vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1290         for (i = 0; i < limit; i++) {
1291                 RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1292                     BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1293                 RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1294                 vrcb += sizeof(struct bge_rcb);
1295         }
1296
1297         /* Configure send ring RCB 0 (we use only the first ring) */
1298         vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1299         BGE_HOSTADDR(taddr, sc->bnx_ldata.bnx_tx_ring_paddr);
1300         RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1301         RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1302         if (BNX_IS_5717_PLUS(sc)) {
1303                 RCB_WRITE_4(sc, vrcb, bge_nicaddr, BGE_SEND_RING_5717);
1304         } else {
1305                 RCB_WRITE_4(sc, vrcb, bge_nicaddr,
1306                     BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1307         }
1308         RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1309             BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1310
1311         /*
1312          * Disable all receive return rings by setting the
1313          * 'ring disabled' bit in the flags field of all the receive
1314          * return ring control blocks, located in NIC memory.
1315          */
1316         if (BNX_IS_5717_PLUS(sc)) {
1317                 /* Should be 17, use 16 until we get an SRAM map. */
1318                 limit = 16;
1319         } else if (BNX_IS_57765_FAMILY(sc) ||
1320             sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
1321                 limit = 4;
1322         } else {
1323                 limit = 1;
1324         }
1325         /* Disable all receive return rings. */
1326         vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1327         for (i = 0; i < limit; i++) {
1328                 RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0);
1329                 RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0);
1330                 RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1331                     BGE_RCB_FLAG_RING_DISABLED);
1332                 RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1333                 bnx_writembx(sc, BGE_MBX_RX_CONS0_LO +
1334                     (i * (sizeof(uint64_t))), 0);
1335                 vrcb += sizeof(struct bge_rcb);
1336         }
1337
1338         /*
1339          * Set up receive return ring 0.  Note that the NIC address
1340          * for RX return rings is 0x0.  The return rings live entirely
1341          * within the host, so the nicaddr field in the RCB isn't used.
1342          */
1343         vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1344         BGE_HOSTADDR(taddr, sc->bnx_ldata.bnx_rx_return_ring_paddr);
1345         RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1346         RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1347         RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1348         RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1349             BGE_RCB_MAXLEN_FLAGS(sc->bnx_return_ring_cnt, 0));
1350
1351         /* Set random backoff seed for TX */
1352         CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1353             sc->arpcom.ac_enaddr[0] + sc->arpcom.ac_enaddr[1] +
1354             sc->arpcom.ac_enaddr[2] + sc->arpcom.ac_enaddr[3] +
1355             sc->arpcom.ac_enaddr[4] + sc->arpcom.ac_enaddr[5] +
1356             BGE_TX_BACKOFF_SEED_MASK);
1357
1358         /* Set inter-packet gap */
1359         val = 0x2620;
1360         if (sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
1361             sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
1362                 val |= CSR_READ_4(sc, BGE_TX_LENGTHS) &
1363                     (BGE_TXLEN_JMB_FRM_LEN_MSK | BGE_TXLEN_CNT_DN_VAL_MSK);
1364         }
1365         CSR_WRITE_4(sc, BGE_TX_LENGTHS, val);
1366
1367         /*
1368          * Specify which ring to use for packets that don't match
1369          * any RX rules.
1370          */
1371         CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1372
1373         /*
1374          * Configure number of RX lists. One interrupt distribution
1375          * list, sixteen active lists, one bad frames class.
1376          */
1377         CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1378
1379         /* Inialize RX list placement stats mask. */
1380         CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1381         CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1382
1383         /* Disable host coalescing until we get it set up */
1384         CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1385
1386         /* Poll to make sure it's shut down. */
1387         for (i = 0; i < BNX_TIMEOUT; i++) {
1388                 if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1389                         break;
1390                 DELAY(10);
1391         }
1392
1393         if (i == BNX_TIMEOUT) {
1394                 if_printf(&sc->arpcom.ac_if,
1395                           "host coalescing engine failed to idle\n");
1396                 return(ENXIO);
1397         }
1398
1399         /* Set up host coalescing defaults */
1400         CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bnx_rx_coal_ticks);
1401         CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bnx_tx_coal_ticks);
1402         CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bnx_rx_coal_bds);
1403         CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bnx_tx_coal_bds);
1404         CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, sc->bnx_rx_coal_bds_int);
1405         CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, sc->bnx_tx_coal_bds_int);
1406
1407         /* Set up address of status block */
1408         bzero(sc->bnx_ldata.bnx_status_block, BGE_STATUS_BLK_SZ);
1409         CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI,
1410             BGE_ADDR_HI(sc->bnx_ldata.bnx_status_block_paddr));
1411         CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO,
1412             BGE_ADDR_LO(sc->bnx_ldata.bnx_status_block_paddr));
1413
1414         /* Set up status block partail update size. */
1415         val = BGE_STATBLKSZ_32BYTE;
1416 #if 0
1417         /*
1418          * Does not seem to have visible effect in both
1419          * bulk data (1472B UDP datagram) and tiny data
1420          * (18B UDP datagram) TX tests.
1421          */
1422         val |= BGE_HCCMODE_CLRTICK_TX;
1423 #endif
1424         /* Turn on host coalescing state machine */
1425         CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
1426
1427         /* Turn on RX BD completion state machine and enable attentions */
1428         CSR_WRITE_4(sc, BGE_RBDC_MODE,
1429             BGE_RBDCMODE_ENABLE|BGE_RBDCMODE_ATTN);
1430
1431         /* Turn on RX list placement state machine */
1432         CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
1433
1434         val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
1435             BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
1436             BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
1437             BGE_MACMODE_FRMHDR_DMA_ENB;
1438
1439         if (sc->bnx_flags & BNX_FLAG_TBI)
1440                 val |= BGE_PORTMODE_TBI;
1441         else if (sc->bnx_flags & BNX_FLAG_MII_SERDES)
1442                 val |= BGE_PORTMODE_GMII;
1443         else
1444                 val |= BGE_PORTMODE_MII;
1445
1446         /* Turn on DMA, clear stats */
1447         CSR_WRITE_4(sc, BGE_MAC_MODE, val);
1448
1449         /* Set misc. local control, enable interrupts on attentions */
1450         CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
1451
1452 #ifdef notdef
1453         /* Assert GPIO pins for PHY reset */
1454         BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
1455             BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
1456         BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
1457             BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
1458 #endif
1459
1460         /* Turn on write DMA state machine */
1461         val = BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
1462         /* Enable host coalescing bug fix. */
1463         val |= BGE_WDMAMODE_STATUS_TAG_FIX;
1464         if (sc->bnx_asicrev == BGE_ASICREV_BCM5785) {
1465                 /* Request larger DMA burst size to get better performance. */
1466                 val |= BGE_WDMAMODE_BURST_ALL_DATA;
1467         }
1468         CSR_WRITE_4(sc, BGE_WDMA_MODE, val);
1469         DELAY(40);
1470
1471         if (BNX_IS_57765_PLUS(sc)) {
1472                 uint32_t dmactl, dmactl_reg;
1473
1474                 if (sc->bnx_asicrev == BGE_ASICREV_BCM5762)
1475                         dmactl_reg = BGE_RDMA_RSRVCTRL2;
1476                 else
1477                         dmactl_reg = BGE_RDMA_RSRVCTRL;
1478
1479                 dmactl = CSR_READ_4(sc, dmactl_reg);
1480                 /*
1481                  * Adjust tx margin to prevent TX data corruption and
1482                  * fix internal FIFO overflow.
1483                  */
1484                 if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 ||
1485                     sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
1486                     sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
1487                         dmactl &= ~(BGE_RDMA_RSRVCTRL_FIFO_LWM_MASK |
1488                             BGE_RDMA_RSRVCTRL_FIFO_HWM_MASK |
1489                             BGE_RDMA_RSRVCTRL_TXMRGN_MASK);
1490                         dmactl |= BGE_RDMA_RSRVCTRL_FIFO_LWM_1_5K |
1491                             BGE_RDMA_RSRVCTRL_FIFO_HWM_1_5K |
1492                             BGE_RDMA_RSRVCTRL_TXMRGN_320B;
1493                 }
1494                 /*
1495                  * Enable fix for read DMA FIFO overruns.
1496                  * The fix is to limit the number of RX BDs
1497                  * the hardware would fetch at a fime.
1498                  */
1499                 CSR_WRITE_4(sc, dmactl_reg,
1500                     dmactl | BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
1501         }
1502
1503         if (sc->bnx_asicrev == BGE_ASICREV_BCM5719) {
1504                 CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
1505                     CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
1506                     BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
1507                     BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
1508         } else if (sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
1509             sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
1510                 uint32_t ctrl_reg;
1511
1512                 if (sc->bnx_asicrev == BGE_ASICREV_BCM5762)
1513                         ctrl_reg = BGE_RDMA_LSO_CRPTEN_CTRL2;
1514                 else
1515                         ctrl_reg = BGE_RDMA_LSO_CRPTEN_CTRL;
1516
1517                 /*
1518                  * Allow 4KB burst length reads for non-LSO frames.
1519                  * Enable 512B burst length reads for buffer descriptors.
1520                  */
1521                 CSR_WRITE_4(sc, ctrl_reg,
1522                     CSR_READ_4(sc, ctrl_reg) |
1523                     BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_512 |
1524                     BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
1525         }
1526
1527         /* Turn on read DMA state machine */
1528         val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
1529         if (sc->bnx_asicrev == BGE_ASICREV_BCM5717)
1530                 val |= BGE_RDMAMODE_MULT_DMA_RD_DIS;
1531         if (sc->bnx_asicrev == BGE_ASICREV_BCM5784 ||
1532             sc->bnx_asicrev == BGE_ASICREV_BCM5785 ||
1533             sc->bnx_asicrev == BGE_ASICREV_BCM57780) {
1534                 val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
1535                     BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
1536                     BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
1537         }
1538         if (sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
1539             sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
1540                 val |= CSR_READ_4(sc, BGE_RDMA_MODE) &
1541                     BGE_RDMAMODE_H2BNC_VLAN_DET;
1542                 /*
1543                  * Allow multiple outstanding read requests from
1544                  * non-LSO read DMA engine.
1545                  */
1546                 val &= ~BGE_RDMAMODE_MULT_DMA_RD_DIS;
1547         }
1548         if (sc->bnx_asicrev == BGE_ASICREV_BCM57766)
1549                 val |= BGE_RDMAMODE_JMB_2K_MMRR;
1550         if (sc->bnx_flags & BNX_FLAG_TSO)
1551                 val |= BGE_RDMAMODE_TSO4_ENABLE;
1552         val |= BGE_RDMAMODE_FIFO_LONG_BURST;
1553         CSR_WRITE_4(sc, BGE_RDMA_MODE, val);
1554         DELAY(40);
1555
1556         /* Turn on RX data completion state machine */
1557         CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
1558
1559         /* Turn on RX BD initiator state machine */
1560         CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
1561
1562         /* Turn on RX data and RX BD initiator state machine */
1563         CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
1564
1565         /* Turn on send BD completion state machine */
1566         CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
1567
1568         /* Turn on send data completion state machine */
1569         val = BGE_SDCMODE_ENABLE;
1570         if (sc->bnx_asicrev == BGE_ASICREV_BCM5761)
1571                 val |= BGE_SDCMODE_CDELAY; 
1572         CSR_WRITE_4(sc, BGE_SDC_MODE, val);
1573
1574         /* Turn on send data initiator state machine */
1575         if (sc->bnx_flags & BNX_FLAG_TSO) {
1576                 CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
1577                     BGE_SDIMODE_HW_LSO_PRE_DMA);
1578         } else {
1579                 CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
1580         }
1581
1582         /* Turn on send BD initiator state machine */
1583         CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
1584
1585         /* Turn on send BD selector state machine */
1586         CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
1587
1588         CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
1589         CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
1590             BGE_SDISTATSCTL_ENABLE|BGE_SDISTATSCTL_FASTER);
1591
1592         /* ack/clear link change events */
1593         CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1594             BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1595             BGE_MACSTAT_LINK_CHANGED);
1596         CSR_WRITE_4(sc, BGE_MI_STS, 0);
1597
1598         /*
1599          * Enable attention when the link has changed state for
1600          * devices that use auto polling.
1601          */
1602         if (sc->bnx_flags & BNX_FLAG_TBI) {
1603                 CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
1604         } else {
1605                 if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
1606                         CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
1607                         DELAY(80);
1608                 }
1609         }
1610
1611         /*
1612          * Clear any pending link state attention.
1613          * Otherwise some link state change events may be lost until attention
1614          * is cleared by bnx_intr() -> bnx_softc.bnx_link_upd() sequence.
1615          * It's not necessary on newer BCM chips - perhaps enabling link
1616          * state change attentions implies clearing pending attention.
1617          */
1618         CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1619             BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1620             BGE_MACSTAT_LINK_CHANGED);
1621
1622         /* Enable link state change attentions. */
1623         BNX_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
1624
1625         return(0);
1626 }
1627
1628 /*
1629  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
1630  * against our list and return its name if we find a match. Note
1631  * that since the Broadcom controller contains VPD support, we
1632  * can get the device name string from the controller itself instead
1633  * of the compiled-in string. This is a little slow, but it guarantees
1634  * we'll always announce the right product name.
1635  */
1636 static int
1637 bnx_probe(device_t dev)
1638 {
1639         const struct bnx_type *t;
1640         uint16_t product, vendor;
1641
1642         if (!pci_is_pcie(dev))
1643                 return ENXIO;
1644
1645         product = pci_get_device(dev);
1646         vendor = pci_get_vendor(dev);
1647
1648         for (t = bnx_devs; t->bnx_name != NULL; t++) {
1649                 if (vendor == t->bnx_vid && product == t->bnx_did)
1650                         break;
1651         }
1652         if (t->bnx_name == NULL)
1653                 return ENXIO;
1654
1655         device_set_desc(dev, t->bnx_name);
1656         return 0;
1657 }
1658
1659 static int
1660 bnx_attach(device_t dev)
1661 {
1662         struct ifnet *ifp;
1663         struct bnx_softc *sc;
1664         uint32_t hwcfg = 0;
1665         int error = 0, rid, capmask;
1666         uint8_t ether_addr[ETHER_ADDR_LEN];
1667         uint16_t product;
1668         driver_intr_t *intr_func;
1669         uintptr_t mii_priv = 0;
1670         u_int intr_flags;
1671 #ifdef BNX_TSO_DEBUG
1672         char desc[32];
1673         int i;
1674 #endif
1675
1676         sc = device_get_softc(dev);
1677         sc->bnx_dev = dev;
1678         callout_init_mp(&sc->bnx_stat_timer);
1679         callout_init_mp(&sc->bnx_intr_timer);
1680         lwkt_serialize_init(&sc->bnx_jslot_serializer);
1681
1682         product = pci_get_device(dev);
1683
1684 #ifndef BURN_BRIDGES
1685         if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1686                 uint32_t irq, mem;
1687
1688                 irq = pci_read_config(dev, PCIR_INTLINE, 4);
1689                 mem = pci_read_config(dev, BGE_PCI_BAR0, 4);
1690
1691                 device_printf(dev, "chip is in D%d power mode "
1692                     "-- setting to D0\n", pci_get_powerstate(dev));
1693
1694                 pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1695
1696                 pci_write_config(dev, PCIR_INTLINE, irq, 4);
1697                 pci_write_config(dev, BGE_PCI_BAR0, mem, 4);
1698         }
1699 #endif  /* !BURN_BRIDGE */
1700
1701         /*
1702          * Map control/status registers.
1703          */
1704         pci_enable_busmaster(dev);
1705
1706         rid = BGE_PCI_BAR0;
1707         sc->bnx_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
1708             RF_ACTIVE);
1709
1710         if (sc->bnx_res == NULL) {
1711                 device_printf(dev, "couldn't map memory\n");
1712                 return ENXIO;
1713         }
1714
1715         sc->bnx_btag = rman_get_bustag(sc->bnx_res);
1716         sc->bnx_bhandle = rman_get_bushandle(sc->bnx_res);
1717
1718         /* Save various chip information */
1719         sc->bnx_chipid =
1720             pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >>
1721             BGE_PCIMISCCTL_ASICREV_SHIFT;
1722         if (BGE_ASICREV(sc->bnx_chipid) == BGE_ASICREV_USE_PRODID_REG) {
1723                 /* All chips having dedicated ASICREV register have CPMU */
1724                 sc->bnx_flags |= BNX_FLAG_CPMU;
1725
1726                 switch (product) {
1727                 case PCI_PRODUCT_BROADCOM_BCM5717:
1728                 case PCI_PRODUCT_BROADCOM_BCM5717C:
1729                 case PCI_PRODUCT_BROADCOM_BCM5718:
1730                 case PCI_PRODUCT_BROADCOM_BCM5719:
1731                 case PCI_PRODUCT_BROADCOM_BCM5720_ALT:
1732                 case PCI_PRODUCT_BROADCOM_BCM5725:
1733                 case PCI_PRODUCT_BROADCOM_BCM5727:
1734                 case PCI_PRODUCT_BROADCOM_BCM5762:
1735                         sc->bnx_chipid = pci_read_config(dev,
1736                             BGE_PCI_GEN2_PRODID_ASICREV, 4);
1737                         break;
1738
1739                 case PCI_PRODUCT_BROADCOM_BCM57761:
1740                 case PCI_PRODUCT_BROADCOM_BCM57762:
1741                 case PCI_PRODUCT_BROADCOM_BCM57765:
1742                 case PCI_PRODUCT_BROADCOM_BCM57766:
1743                 case PCI_PRODUCT_BROADCOM_BCM57781:
1744                 case PCI_PRODUCT_BROADCOM_BCM57782:
1745                 case PCI_PRODUCT_BROADCOM_BCM57785:
1746                 case PCI_PRODUCT_BROADCOM_BCM57786:
1747                 case PCI_PRODUCT_BROADCOM_BCM57791:
1748                 case PCI_PRODUCT_BROADCOM_BCM57795:
1749                         sc->bnx_chipid = pci_read_config(dev,
1750                             BGE_PCI_GEN15_PRODID_ASICREV, 4);
1751                         break;
1752
1753                 default:
1754                         sc->bnx_chipid = pci_read_config(dev,
1755                             BGE_PCI_PRODID_ASICREV, 4);
1756                         break;
1757                 }
1758         }
1759         if (sc->bnx_chipid == BGE_CHIPID_BCM5717_C0)
1760                 sc->bnx_chipid = BGE_CHIPID_BCM5720_A0;
1761
1762         sc->bnx_asicrev = BGE_ASICREV(sc->bnx_chipid);
1763         sc->bnx_chiprev = BGE_CHIPREV(sc->bnx_chipid);
1764
1765         switch (sc->bnx_asicrev) {
1766         case BGE_ASICREV_BCM5717:
1767         case BGE_ASICREV_BCM5719:
1768         case BGE_ASICREV_BCM5720:
1769                 sc->bnx_flags |= BNX_FLAG_5717_PLUS | BNX_FLAG_57765_PLUS;
1770                 break;
1771
1772         case BGE_ASICREV_BCM5762:
1773                 sc->bnx_flags |= BNX_FLAG_57765_PLUS;
1774                 break;
1775
1776         case BGE_ASICREV_BCM57765:
1777         case BGE_ASICREV_BCM57766:
1778                 sc->bnx_flags |= BNX_FLAG_57765_FAMILY | BNX_FLAG_57765_PLUS;
1779                 break;
1780         }
1781         sc->bnx_flags |= BNX_FLAG_SHORTDMA;
1782
1783         sc->bnx_flags |= BNX_FLAG_TSO;
1784         if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 &&
1785             sc->bnx_chipid == BGE_CHIPID_BCM5719_A0)
1786                 sc->bnx_flags &= ~BNX_FLAG_TSO;
1787
1788         if (sc->bnx_asicrev == BGE_ASICREV_BCM5717 ||
1789             BNX_IS_57765_FAMILY(sc)) {
1790                 /*
1791                  * All BCM57785 and BCM5718 families chips have a bug that
1792                  * under certain situation interrupt will not be enabled
1793                  * even if status tag is written to BGE_MBX_IRQ0_LO mailbox.
1794                  *
1795                  * While BCM5719 and BCM5720 have a hardware workaround
1796                  * which could fix the above bug.
1797                  * See the comment near BGE_PCIDMARWCTL_TAGGED_STATUS_WA in
1798                  * bnx_chipinit().
1799                  *
1800                  * For the rest of the chips in these two families, we will
1801                  * have to poll the status block at high rate (10ms currently)
1802                  * to check whether the interrupt is hosed or not.
1803                  * See bnx_intr_check() for details.
1804                  */
1805                 sc->bnx_flags |= BNX_FLAG_STATUSTAG_BUG;
1806         }
1807
1808         sc->bnx_pciecap = pci_get_pciecap_ptr(sc->bnx_dev);
1809         if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 ||
1810             sc->bnx_asicrev == BGE_ASICREV_BCM5720)
1811                 pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_2048);
1812         else
1813                 pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_4096);
1814         device_printf(dev, "CHIP ID 0x%08x; "
1815                       "ASIC REV 0x%02x; CHIP REV 0x%02x\n",
1816                       sc->bnx_chipid, sc->bnx_asicrev, sc->bnx_chiprev);
1817
1818         /*
1819          * Set various PHY quirk flags.
1820          */
1821
1822         capmask = MII_CAPMASK_DEFAULT;
1823         if (product == PCI_PRODUCT_BROADCOM_BCM57791 ||
1824             product == PCI_PRODUCT_BROADCOM_BCM57795) {
1825                 /* 10/100 only */
1826                 capmask &= ~BMSR_EXTSTAT;
1827         }
1828
1829         mii_priv |= BRGPHY_FLAG_WIRESPEED;
1830         if (sc->bnx_chipid == BGE_CHIPID_BCM5762_A0)
1831                 mii_priv |= BRGPHY_FLAG_5762_A0;
1832
1833         /* Initialize if_name earlier, so if_printf could be used */
1834         ifp = &sc->arpcom.ac_if;
1835         if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1836
1837         /* Try to reset the chip. */
1838         bnx_reset(sc);
1839
1840         if (bnx_chipinit(sc)) {
1841                 device_printf(dev, "chip initialization failed\n");
1842                 error = ENXIO;
1843                 goto fail;
1844         }
1845
1846         /*
1847          * Get station address
1848          */
1849         error = bnx_get_eaddr(sc, ether_addr);
1850         if (error) {
1851                 device_printf(dev, "failed to read station address\n");
1852                 goto fail;
1853         }
1854
1855         if (BNX_IS_57765_PLUS(sc)) {
1856                 sc->bnx_return_ring_cnt = BGE_RETURN_RING_CNT;
1857         } else {
1858                 /* 5705/5750 limits RX return ring to 512 entries. */
1859                 sc->bnx_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
1860         }
1861
1862         error = bnx_dma_alloc(sc);
1863         if (error)
1864                 goto fail;
1865
1866         /*
1867          * Allocate interrupt
1868          */
1869         sc->bnx_irq_type = pci_alloc_1intr(dev, bnx_msi_enable, &sc->bnx_irq_rid,
1870             &intr_flags);
1871
1872         sc->bnx_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->bnx_irq_rid,
1873             intr_flags);
1874         if (sc->bnx_irq == NULL) {
1875                 device_printf(dev, "couldn't map interrupt\n");
1876                 error = ENXIO;
1877                 goto fail;
1878         }
1879
1880         if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI) {
1881                 sc->bnx_flags |= BNX_FLAG_ONESHOT_MSI;
1882                 bnx_enable_msi(sc);
1883         }
1884
1885         /* Set default tuneable values. */
1886         sc->bnx_rx_coal_ticks = BNX_RX_COAL_TICKS_DEF;
1887         sc->bnx_tx_coal_ticks = BNX_TX_COAL_TICKS_DEF;
1888         sc->bnx_rx_coal_bds = BNX_RX_COAL_BDS_DEF;
1889         sc->bnx_tx_coal_bds = BNX_TX_COAL_BDS_DEF;
1890         sc->bnx_rx_coal_bds_int = BNX_RX_COAL_BDS_INT_DEF;
1891         sc->bnx_tx_coal_bds_int = BNX_TX_COAL_BDS_INT_DEF;
1892         sc->bnx_tx_wreg = BNX_TX_WREG_NSEGS;
1893
1894         /* Set up ifnet structure */
1895         ifp->if_softc = sc;
1896         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1897         ifp->if_ioctl = bnx_ioctl;
1898         ifp->if_start = bnx_start;
1899 #ifdef IFPOLL_ENABLE
1900         ifp->if_npoll = bnx_npoll;
1901 #endif
1902         ifp->if_watchdog = bnx_watchdog;
1903         ifp->if_init = bnx_init;
1904         ifp->if_mtu = ETHERMTU;
1905         ifp->if_capabilities = IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
1906         ifq_set_maxlen(&ifp->if_snd, BGE_TX_RING_CNT - 1);
1907         ifq_set_ready(&ifp->if_snd);
1908
1909         ifp->if_capabilities |= IFCAP_HWCSUM;
1910         ifp->if_hwassist = BNX_CSUM_FEATURES;
1911         if (sc->bnx_flags & BNX_FLAG_TSO) {
1912                 ifp->if_capabilities |= IFCAP_TSO;
1913                 ifp->if_hwassist |= CSUM_TSO;
1914         }
1915         ifp->if_capenable = ifp->if_capabilities;
1916
1917         /*
1918          * Figure out what sort of media we have by checking the
1919          * hardware config word in the first 32k of NIC internal memory,
1920          * or fall back to examining the EEPROM if necessary.
1921          * Note: on some BCM5700 cards, this value appears to be unset.
1922          * If that's the case, we have to rely on identifying the NIC
1923          * by its PCI subsystem ID, as we do below for the SysKonnect
1924          * SK-9D41.
1925          */
1926         if (bnx_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
1927                 hwcfg = bnx_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
1928         } else {
1929                 if (bnx_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET,
1930                                     sizeof(hwcfg))) {
1931                         device_printf(dev, "failed to read EEPROM\n");
1932                         error = ENXIO;
1933                         goto fail;
1934                 }
1935                 hwcfg = ntohl(hwcfg);
1936         }
1937
1938         /* The SysKonnect SK-9D41 is a 1000baseSX card. */
1939         if (pci_get_subvendor(dev) == PCI_PRODUCT_SCHNEIDERKOCH_SK_9D41 ||
1940             (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER)
1941                 sc->bnx_flags |= BNX_FLAG_TBI;
1942
1943         /* Setup MI MODE */
1944         if (sc->bnx_flags & BNX_FLAG_CPMU)
1945                 sc->bnx_mi_mode = BGE_MIMODE_500KHZ_CONST;
1946         else
1947                 sc->bnx_mi_mode = BGE_MIMODE_BASE;
1948
1949         /* Setup link status update stuffs */
1950         if (sc->bnx_flags & BNX_FLAG_TBI) {
1951                 sc->bnx_link_upd = bnx_tbi_link_upd;
1952                 sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
1953         } else if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
1954                 sc->bnx_link_upd = bnx_autopoll_link_upd;
1955                 sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
1956         } else {
1957                 sc->bnx_link_upd = bnx_copper_link_upd;
1958                 sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
1959         }
1960
1961         /* Set default PHY address */
1962         sc->bnx_phyno = 1;
1963
1964         /*
1965          * PHY address mapping for various devices.
1966          *
1967          *          | F0 Cu | F0 Sr | F1 Cu | F1 Sr |
1968          * ---------+-------+-------+-------+-------+
1969          * BCM57XX  |   1   |   X   |   X   |   X   |
1970          * BCM5704  |   1   |   X   |   1   |   X   |
1971          * BCM5717  |   1   |   8   |   2   |   9   |
1972          * BCM5719  |   1   |   8   |   2   |   9   |
1973          * BCM5720  |   1   |   8   |   2   |   9   |
1974          *
1975          * Other addresses may respond but they are not
1976          * IEEE compliant PHYs and should be ignored.
1977          */
1978         if (BNX_IS_5717_PLUS(sc)) {
1979                 int f;
1980
1981                 f = pci_get_function(dev);
1982                 if (sc->bnx_chipid == BGE_CHIPID_BCM5717_A0) {
1983                         if (CSR_READ_4(sc, BGE_SGDIG_STS) &
1984                             BGE_SGDIGSTS_IS_SERDES)
1985                                 sc->bnx_phyno = f + 8;
1986                         else
1987                                 sc->bnx_phyno = f + 1;
1988                 } else {
1989                         if (CSR_READ_4(sc, BGE_CPMU_PHY_STRAP) &
1990                             BGE_CPMU_PHY_STRAP_IS_SERDES)
1991                                 sc->bnx_phyno = f + 8;
1992                         else
1993                                 sc->bnx_phyno = f + 1;
1994                 }
1995         }
1996
1997         if (sc->bnx_flags & BNX_FLAG_TBI) {
1998                 ifmedia_init(&sc->bnx_ifmedia, IFM_IMASK,
1999                     bnx_ifmedia_upd, bnx_ifmedia_sts);
2000                 ifmedia_add(&sc->bnx_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2001                 ifmedia_add(&sc->bnx_ifmedia,
2002                     IFM_ETHER|IFM_1000_SX|IFM_FDX, 0, NULL);
2003                 ifmedia_add(&sc->bnx_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2004                 ifmedia_set(&sc->bnx_ifmedia, IFM_ETHER|IFM_AUTO);
2005                 sc->bnx_ifmedia.ifm_media = sc->bnx_ifmedia.ifm_cur->ifm_media;
2006         } else {
2007                 struct mii_probe_args mii_args;
2008
2009                 mii_probe_args_init(&mii_args, bnx_ifmedia_upd, bnx_ifmedia_sts);
2010                 mii_args.mii_probemask = 1 << sc->bnx_phyno;
2011                 mii_args.mii_capmask = capmask;
2012                 mii_args.mii_privtag = MII_PRIVTAG_BRGPHY;
2013                 mii_args.mii_priv = mii_priv;
2014
2015                 error = mii_probe(dev, &sc->bnx_miibus, &mii_args);
2016                 if (error) {
2017                         device_printf(dev, "MII without any PHY!\n");
2018                         goto fail;
2019                 }
2020         }
2021
2022         /*
2023          * Create sysctl nodes.
2024          */
2025         sysctl_ctx_init(&sc->bnx_sysctl_ctx);
2026         sc->bnx_sysctl_tree = SYSCTL_ADD_NODE(&sc->bnx_sysctl_ctx,
2027                                               SYSCTL_STATIC_CHILDREN(_hw),
2028                                               OID_AUTO,
2029                                               device_get_nameunit(dev),
2030                                               CTLFLAG_RD, 0, "");
2031         if (sc->bnx_sysctl_tree == NULL) {
2032                 device_printf(dev, "can't add sysctl node\n");
2033                 error = ENXIO;
2034                 goto fail;
2035         }
2036
2037         SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2038                         SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2039                         OID_AUTO, "rx_coal_ticks",
2040                         CTLTYPE_INT | CTLFLAG_RW,
2041                         sc, 0, bnx_sysctl_rx_coal_ticks, "I",
2042                         "Receive coalescing ticks (usec).");
2043         SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2044                         SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2045                         OID_AUTO, "tx_coal_ticks",
2046                         CTLTYPE_INT | CTLFLAG_RW,
2047                         sc, 0, bnx_sysctl_tx_coal_ticks, "I",
2048                         "Transmit coalescing ticks (usec).");
2049         SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2050                         SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2051                         OID_AUTO, "rx_coal_bds",
2052                         CTLTYPE_INT | CTLFLAG_RW,
2053                         sc, 0, bnx_sysctl_rx_coal_bds, "I",
2054                         "Receive max coalesced BD count.");
2055         SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2056                         SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2057                         OID_AUTO, "tx_coal_bds",
2058                         CTLTYPE_INT | CTLFLAG_RW,
2059                         sc, 0, bnx_sysctl_tx_coal_bds, "I",
2060                         "Transmit max coalesced BD count.");
2061         /*
2062          * A common design characteristic for many Broadcom
2063          * client controllers is that they only support a
2064          * single outstanding DMA read operation on the PCIe
2065          * bus. This means that it will take twice as long to
2066          * fetch a TX frame that is split into header and
2067          * payload buffers as it does to fetch a single,
2068          * contiguous TX frame (2 reads vs. 1 read). For these
2069          * controllers, coalescing buffers to reduce the number
2070          * of memory reads is effective way to get maximum
2071          * performance(about 940Mbps).  Without collapsing TX
2072          * buffers the maximum TCP bulk transfer performance
2073          * is about 850Mbps. However forcing coalescing mbufs
2074          * consumes a lot of CPU cycles, so leave it off by
2075          * default.
2076          */
2077         SYSCTL_ADD_INT(&sc->bnx_sysctl_ctx,
2078             SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2079             "force_defrag", CTLFLAG_RW, &sc->bnx_force_defrag, 0,
2080             "Force defragment on TX path");
2081
2082         SYSCTL_ADD_INT(&sc->bnx_sysctl_ctx,
2083             SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2084             "tx_wreg", CTLFLAG_RW, &sc->bnx_tx_wreg, 0,
2085             "# of segments before writing to hardware register");
2086
2087         SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2088             SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2089             "rx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2090             sc, 0, bnx_sysctl_rx_coal_bds_int, "I",
2091             "Receive max coalesced BD count during interrupt.");
2092         SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2093             SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2094             "tx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2095             sc, 0, bnx_sysctl_tx_coal_bds_int, "I",
2096             "Transmit max coalesced BD count during interrupt.");
2097
2098 #ifdef BNX_TSO_DEBUG
2099         for (i = 0; i < BNX_TSO_NSTATS; ++i) {
2100                 ksnprintf(desc, sizeof(desc), "tso%d", i + 1);
2101                 SYSCTL_ADD_ULONG(&sc->bnx_sysctl_ctx,
2102                     SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2103                     desc, CTLFLAG_RW, &sc->bnx_tsosegs[i], "");
2104         }
2105 #endif
2106
2107         /*
2108          * Call MI attach routine.
2109          */
2110         ether_ifattach(ifp, ether_addr, NULL);
2111
2112         ifq_set_cpuid(&ifp->if_snd, sc->bnx_intr_cpuid);
2113
2114 #ifdef IFPOLL_ENABLE
2115         ifpoll_compat_setup(&sc->bnx_npoll,
2116             &sc->bnx_sysctl_ctx, sc->bnx_sysctl_tree,
2117             device_get_unit(dev), ifp->if_serializer);
2118 #endif
2119
2120         if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI) {
2121                 if (sc->bnx_flags & BNX_FLAG_ONESHOT_MSI) {
2122                         intr_func = bnx_msi_oneshot;
2123                         if (bootverbose)
2124                                 device_printf(dev, "oneshot MSI\n");
2125                 } else {
2126                         intr_func = bnx_msi;
2127                 }
2128         } else {
2129                 intr_func = bnx_intr_legacy;
2130         }
2131         error = bus_setup_intr(dev, sc->bnx_irq, INTR_MPSAFE, intr_func, sc,
2132             &sc->bnx_intrhand, ifp->if_serializer);
2133         if (error) {
2134                 ether_ifdetach(ifp);
2135                 device_printf(dev, "couldn't set up irq\n");
2136                 goto fail;
2137         }
2138
2139         sc->bnx_intr_cpuid = rman_get_cpuid(sc->bnx_irq);
2140         sc->bnx_stat_cpuid = sc->bnx_intr_cpuid;
2141
2142         return(0);
2143 fail:
2144         bnx_detach(dev);
2145         return(error);
2146 }
2147
2148 static int
2149 bnx_detach(device_t dev)
2150 {
2151         struct bnx_softc *sc = device_get_softc(dev);
2152
2153         if (device_is_attached(dev)) {
2154                 struct ifnet *ifp = &sc->arpcom.ac_if;
2155
2156                 lwkt_serialize_enter(ifp->if_serializer);
2157                 bnx_stop(sc);
2158                 bnx_reset(sc);
2159                 bus_teardown_intr(dev, sc->bnx_irq, sc->bnx_intrhand);
2160                 lwkt_serialize_exit(ifp->if_serializer);
2161
2162                 ether_ifdetach(ifp);
2163         }
2164
2165         if (sc->bnx_flags & BNX_FLAG_TBI)
2166                 ifmedia_removeall(&sc->bnx_ifmedia);
2167         if (sc->bnx_miibus)
2168                 device_delete_child(dev, sc->bnx_miibus);
2169         bus_generic_detach(dev);
2170
2171         if (sc->bnx_irq != NULL) {
2172                 bus_release_resource(dev, SYS_RES_IRQ, sc->bnx_irq_rid,
2173                     sc->bnx_irq);
2174         }
2175         if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI)
2176                 pci_release_msi(dev);
2177
2178         if (sc->bnx_res != NULL) {
2179                 bus_release_resource(dev, SYS_RES_MEMORY,
2180                     BGE_PCI_BAR0, sc->bnx_res);
2181         }
2182
2183         if (sc->bnx_sysctl_tree != NULL)
2184                 sysctl_ctx_free(&sc->bnx_sysctl_ctx);
2185
2186         bnx_dma_free(sc);
2187
2188         return 0;
2189 }
2190
2191 static void
2192 bnx_reset(struct bnx_softc *sc)
2193 {
2194         device_t dev;
2195         uint32_t cachesize, command, pcistate, reset;
2196         void (*write_op)(struct bnx_softc *, uint32_t, uint32_t);
2197         int i, val = 0;
2198         uint16_t devctl;
2199
2200         dev = sc->bnx_dev;
2201
2202         write_op = bnx_writemem_direct;
2203
2204         /* Save some important PCI state. */
2205         cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4);
2206         command = pci_read_config(dev, BGE_PCI_CMD, 4);
2207         pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4);
2208
2209         pci_write_config(dev, BGE_PCI_MISC_CTL,
2210             BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2211             BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2212             BGE_PCIMISCCTL_TAGGED_STATUS, 4);
2213
2214         /* Disable fastboot on controllers that support it. */
2215         if (bootverbose)
2216                 if_printf(&sc->arpcom.ac_if, "Disabling fastboot\n");
2217         CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0);
2218
2219         /*
2220          * Write the magic number to SRAM at offset 0xB50.
2221          * When firmware finishes its initialization it will
2222          * write ~BGE_MAGIC_NUMBER to the same location.
2223          */
2224         bnx_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2225
2226         reset = BGE_MISCCFG_RESET_CORE_CLOCKS|(65<<1);
2227
2228         /* XXX: Broadcom Linux driver. */
2229         /* Force PCI-E 1.0a mode */
2230         if (!BNX_IS_57765_PLUS(sc) &&
2231             CSR_READ_4(sc, BGE_PCIE_PHY_TSTCTL) ==
2232             (BGE_PCIE_PHY_TSTCTL_PSCRAM |
2233              BGE_PCIE_PHY_TSTCTL_PCIE10)) {
2234                 CSR_WRITE_4(sc, BGE_PCIE_PHY_TSTCTL,
2235                     BGE_PCIE_PHY_TSTCTL_PSCRAM);
2236         }
2237         if (sc->bnx_chipid != BGE_CHIPID_BCM5750_A0) {
2238                 /* Prevent PCIE link training during global reset */
2239                 CSR_WRITE_4(sc, BGE_MISC_CFG, (1<<29));
2240                 reset |= (1<<29);
2241         }
2242
2243         /* 
2244          * Set GPHY Power Down Override to leave GPHY
2245          * powered up in D0 uninitialized.
2246          */
2247         if ((sc->bnx_flags & BNX_FLAG_CPMU) == 0)
2248                 reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
2249
2250         /* Issue global reset */
2251         write_op(sc, BGE_MISC_CFG, reset);
2252
2253         DELAY(1000);
2254
2255         /* XXX: Broadcom Linux driver. */
2256         if (sc->bnx_chipid == BGE_CHIPID_BCM5750_A0) {
2257                 uint32_t v;
2258
2259                 DELAY(500000); /* wait for link training to complete */
2260                 v = pci_read_config(dev, 0xc4, 4);
2261                 pci_write_config(dev, 0xc4, v | (1<<15), 4);
2262         }
2263
2264         devctl = pci_read_config(dev, sc->bnx_pciecap + PCIER_DEVCTRL, 2);
2265
2266         /* Disable no snoop and disable relaxed ordering. */
2267         devctl &= ~(PCIEM_DEVCTL_RELAX_ORDER | PCIEM_DEVCTL_NOSNOOP);
2268
2269         /* Old PCI-E chips only support 128 bytes Max PayLoad Size. */
2270         if ((sc->bnx_flags & BNX_FLAG_CPMU) == 0) {
2271                 devctl &= ~PCIEM_DEVCTL_MAX_PAYLOAD_MASK;
2272                 devctl |= PCIEM_DEVCTL_MAX_PAYLOAD_128;
2273         }
2274
2275         pci_write_config(dev, sc->bnx_pciecap + PCIER_DEVCTRL,
2276             devctl, 2);
2277
2278         /* Clear error status. */
2279         pci_write_config(dev, sc->bnx_pciecap + PCIER_DEVSTS,
2280             PCIEM_DEVSTS_CORR_ERR |
2281             PCIEM_DEVSTS_NFATAL_ERR |
2282             PCIEM_DEVSTS_FATAL_ERR |
2283             PCIEM_DEVSTS_UNSUPP_REQ, 2);
2284
2285         /* Reset some of the PCI state that got zapped by reset */
2286         pci_write_config(dev, BGE_PCI_MISC_CTL,
2287             BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2288             BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2289             BGE_PCIMISCCTL_TAGGED_STATUS, 4);
2290         pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4);
2291         pci_write_config(dev, BGE_PCI_CMD, command, 4);
2292         write_op(sc, BGE_MISC_CFG, (65 << 1));
2293
2294         /* Enable memory arbiter */
2295         CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
2296
2297         /*
2298          * Poll until we see the 1's complement of the magic number.
2299          * This indicates that the firmware initialization is complete.
2300          */
2301         for (i = 0; i < BNX_FIRMWARE_TIMEOUT; i++) {
2302                 val = bnx_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2303                 if (val == ~BGE_MAGIC_NUMBER)
2304                         break;
2305                 DELAY(10);
2306         }
2307         if (i == BNX_FIRMWARE_TIMEOUT) {
2308                 if_printf(&sc->arpcom.ac_if, "firmware handshake "
2309                           "timed out, found 0x%08x\n", val);
2310         }
2311
2312         /* BCM57765 A0 needs additional time before accessing. */
2313         if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0)
2314                 DELAY(10 * 1000);
2315
2316         /*
2317          * XXX Wait for the value of the PCISTATE register to
2318          * return to its original pre-reset state. This is a
2319          * fairly good indicator of reset completion. If we don't
2320          * wait for the reset to fully complete, trying to read
2321          * from the device's non-PCI registers may yield garbage
2322          * results.
2323          */
2324         for (i = 0; i < BNX_TIMEOUT; i++) {
2325                 if (pci_read_config(dev, BGE_PCI_PCISTATE, 4) == pcistate)
2326                         break;
2327                 DELAY(10);
2328         }
2329
2330         /* Fix up byte swapping */
2331         CSR_WRITE_4(sc, BGE_MODE_CTL, bnx_dma_swap_options(sc));
2332
2333         CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2334
2335         /*
2336          * The 5704 in TBI mode apparently needs some special
2337          * adjustment to insure the SERDES drive level is set
2338          * to 1.2V.
2339          */
2340         if (sc->bnx_asicrev == BGE_ASICREV_BCM5704 &&
2341             (sc->bnx_flags & BNX_FLAG_TBI)) {
2342                 uint32_t serdescfg;
2343
2344                 serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2345                 serdescfg = (serdescfg & ~0xFFF) | 0x880;
2346                 CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2347         }
2348
2349         CSR_WRITE_4(sc, BGE_MI_MODE,
2350             sc->bnx_mi_mode & ~BGE_MIMODE_AUTOPOLL);
2351         DELAY(80);
2352
2353         /* XXX: Broadcom Linux driver. */
2354         if (!BNX_IS_57765_PLUS(sc)) {
2355                 uint32_t v;
2356
2357                 /* Enable Data FIFO protection. */
2358                 v = CSR_READ_4(sc, BGE_PCIE_TLDLPL_PORT);
2359                 CSR_WRITE_4(sc, BGE_PCIE_TLDLPL_PORT, v | (1 << 25));
2360         }
2361
2362         DELAY(10000);
2363
2364         if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
2365                 BNX_CLRBIT(sc, BGE_CPMU_CLCK_ORIDE,
2366                     CPMU_CLCK_ORIDE_MAC_ORIDE_EN);
2367         }
2368 }
2369
2370 /*
2371  * Frame reception handling. This is called if there's a frame
2372  * on the receive return list.
2373  *
2374  * Note: we have to be able to handle two possibilities here:
2375  * 1) the frame is from the jumbo recieve ring
2376  * 2) the frame is from the standard receive ring
2377  */
2378
2379 static void
2380 bnx_rxeof(struct bnx_softc *sc, uint16_t rx_prod, int count)
2381 {
2382         struct ifnet *ifp;
2383         int stdcnt = 0, jumbocnt = 0;
2384
2385         ifp = &sc->arpcom.ac_if;
2386
2387         while (sc->bnx_rx_saved_considx != rx_prod && count != 0) {
2388                 struct bge_rx_bd        *cur_rx;
2389                 uint32_t                rxidx;
2390                 struct mbuf             *m = NULL;
2391                 uint16_t                vlan_tag = 0;
2392                 int                     have_tag = 0;
2393
2394                 --count;
2395
2396                 cur_rx =
2397             &sc->bnx_ldata.bnx_rx_return_ring[sc->bnx_rx_saved_considx];
2398
2399                 rxidx = cur_rx->bge_idx;
2400                 BNX_INC(sc->bnx_rx_saved_considx, sc->bnx_return_ring_cnt);
2401
2402                 if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
2403                         have_tag = 1;
2404                         vlan_tag = cur_rx->bge_vlan_tag;
2405                 }
2406
2407                 if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
2408                         BNX_INC(sc->bnx_jumbo, BGE_JUMBO_RX_RING_CNT);
2409                         jumbocnt++;
2410
2411                         if (rxidx != sc->bnx_jumbo) {
2412                                 IFNET_STAT_INC(ifp, ierrors, 1);
2413                                 if_printf(ifp, "sw jumbo index(%d) "
2414                                     "and hw jumbo index(%d) mismatch, drop!\n",
2415                                     sc->bnx_jumbo, rxidx);
2416                                 bnx_setup_rxdesc_jumbo(sc, rxidx);
2417                                 continue;
2418                         }
2419
2420                         m = sc->bnx_cdata.bnx_rx_jumbo_chain[rxidx].bnx_mbuf;
2421                         if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2422                                 IFNET_STAT_INC(ifp, ierrors, 1);
2423                                 bnx_setup_rxdesc_jumbo(sc, sc->bnx_jumbo);
2424                                 continue;
2425                         }
2426                         if (bnx_newbuf_jumbo(sc, sc->bnx_jumbo, 0)) {
2427                                 IFNET_STAT_INC(ifp, ierrors, 1);
2428                                 bnx_setup_rxdesc_jumbo(sc, sc->bnx_jumbo);
2429                                 continue;
2430                         }
2431                 } else {
2432                         BNX_INC(sc->bnx_std, BGE_STD_RX_RING_CNT);
2433                         stdcnt++;
2434
2435                         if (rxidx != sc->bnx_std) {
2436                                 IFNET_STAT_INC(ifp, ierrors, 1);
2437                                 if_printf(ifp, "sw std index(%d) "
2438                                     "and hw std index(%d) mismatch, drop!\n",
2439                                     sc->bnx_std, rxidx);
2440                                 bnx_setup_rxdesc_std(sc, rxidx);
2441                                 continue;
2442                         }
2443
2444                         m = sc->bnx_cdata.bnx_rx_std_chain[rxidx].bnx_mbuf;
2445                         if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2446                                 IFNET_STAT_INC(ifp, ierrors, 1);
2447                                 bnx_setup_rxdesc_std(sc, sc->bnx_std);
2448                                 continue;
2449                         }
2450                         if (bnx_newbuf_std(sc, sc->bnx_std, 0)) {
2451                                 IFNET_STAT_INC(ifp, ierrors, 1);
2452                                 bnx_setup_rxdesc_std(sc, sc->bnx_std);
2453                                 continue;
2454                         }
2455                 }
2456
2457                 IFNET_STAT_INC(ifp, ipackets, 1);
2458                 m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
2459                 m->m_pkthdr.rcvif = ifp;
2460
2461                 if ((ifp->if_capenable & IFCAP_RXCSUM) &&
2462                     (cur_rx->bge_flags & BGE_RXBDFLAG_IPV6) == 0) {
2463                         if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
2464                                 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2465                                 if ((cur_rx->bge_error_flag &
2466                                     BGE_RXERRFLAG_IP_CSUM_NOK) == 0)
2467                                         m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2468                         }
2469                         if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) {
2470                                 m->m_pkthdr.csum_data =
2471                                     cur_rx->bge_tcp_udp_csum;
2472                                 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2473                                     CSUM_PSEUDO_HDR;
2474                         }
2475                 }
2476
2477                 /*
2478                  * If we received a packet with a vlan tag, pass it
2479                  * to vlan_input() instead of ether_input().
2480                  */
2481                 if (have_tag) {
2482                         m->m_flags |= M_VLANTAG;
2483                         m->m_pkthdr.ether_vlantag = vlan_tag;
2484                 }
2485                 ifp->if_input(ifp, m);
2486         }
2487
2488         bnx_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bnx_rx_saved_considx);
2489         if (stdcnt)
2490                 bnx_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bnx_std);
2491         if (jumbocnt)
2492                 bnx_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bnx_jumbo);
2493 }
2494
2495 static void
2496 bnx_txeof(struct bnx_softc *sc, uint16_t tx_cons)
2497 {
2498         struct ifnet *ifp;
2499
2500         ifp = &sc->arpcom.ac_if;
2501
2502         /*
2503          * Go through our tx ring and free mbufs for those
2504          * frames that have been sent.
2505          */
2506         while (sc->bnx_tx_saved_considx != tx_cons) {
2507                 uint32_t idx = 0;
2508
2509                 idx = sc->bnx_tx_saved_considx;
2510                 if (sc->bnx_cdata.bnx_tx_chain[idx] != NULL) {
2511                         IFNET_STAT_INC(ifp, opackets, 1);
2512                         bus_dmamap_unload(sc->bnx_cdata.bnx_tx_mtag,
2513                             sc->bnx_cdata.bnx_tx_dmamap[idx]);
2514                         m_freem(sc->bnx_cdata.bnx_tx_chain[idx]);
2515                         sc->bnx_cdata.bnx_tx_chain[idx] = NULL;
2516                 }
2517                 sc->bnx_txcnt--;
2518                 BNX_INC(sc->bnx_tx_saved_considx, BGE_TX_RING_CNT);
2519         }
2520
2521         if ((BGE_TX_RING_CNT - sc->bnx_txcnt) >=
2522             (BNX_NSEG_RSVD + BNX_NSEG_SPARE))
2523                 ifq_clr_oactive(&ifp->if_snd);
2524
2525         if (sc->bnx_txcnt == 0)
2526                 ifp->if_timer = 0;
2527
2528         if (!ifq_is_empty(&ifp->if_snd))
2529                 if_devstart(ifp);
2530 }
2531
2532 #ifdef IFPOLL_ENABLE
2533
2534 static void
2535 bnx_npoll(struct ifnet *ifp, struct ifpoll_info *info)
2536 {
2537         struct bnx_softc *sc = ifp->if_softc;
2538
2539         ASSERT_SERIALIZED(ifp->if_serializer);
2540
2541         if (info != NULL) {
2542                 int cpuid = sc->bnx_npoll.ifpc_cpuid;
2543
2544                 info->ifpi_rx[cpuid].poll_func = bnx_npoll_compat;
2545                 info->ifpi_rx[cpuid].arg = NULL;
2546                 info->ifpi_rx[cpuid].serializer = ifp->if_serializer;
2547
2548                 if (ifp->if_flags & IFF_RUNNING)
2549                         bnx_disable_intr(sc);
2550                 ifq_set_cpuid(&ifp->if_snd, cpuid);
2551         } else {
2552                 if (ifp->if_flags & IFF_RUNNING)
2553                         bnx_enable_intr(sc);
2554                 ifq_set_cpuid(&ifp->if_snd, sc->bnx_intr_cpuid);
2555         }
2556 }
2557
2558 static void
2559 bnx_npoll_compat(struct ifnet *ifp, void *arg __unused, int cycle)
2560 {
2561         struct bnx_softc *sc = ifp->if_softc;
2562         struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
2563         uint16_t rx_prod, tx_cons;
2564
2565         ASSERT_SERIALIZED(ifp->if_serializer);
2566
2567         if (sc->bnx_npoll.ifpc_stcount-- == 0) {
2568                 sc->bnx_npoll.ifpc_stcount = sc->bnx_npoll.ifpc_stfrac;
2569                 /*
2570                  * Process link state changes.
2571                  */
2572                 bnx_link_poll(sc);
2573         }
2574
2575         sc->bnx_status_tag = sblk->bge_status_tag;
2576
2577         /*
2578          * Use a load fence to ensure that status_tag is saved
2579          * before rx_prod and tx_cons.
2580          */
2581         cpu_lfence();
2582
2583         rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
2584         tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
2585
2586         if (sc->bnx_rx_saved_considx != rx_prod)
2587                 bnx_rxeof(sc, rx_prod, cycle);
2588
2589         if (sc->bnx_tx_saved_considx != tx_cons)
2590                 bnx_txeof(sc, tx_cons);
2591
2592         if (sc->bnx_coal_chg)
2593                 bnx_coal_change(sc);
2594 }
2595
2596 #endif  /* IFPOLL_ENABLE */
2597
2598 static void
2599 bnx_intr_legacy(void *xsc)
2600 {
2601         struct bnx_softc *sc = xsc;
2602         struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
2603
2604         if (sc->bnx_status_tag == sblk->bge_status_tag) {
2605                 uint32_t val;
2606
2607                 val = pci_read_config(sc->bnx_dev, BGE_PCI_PCISTATE, 4);
2608                 if (val & BGE_PCISTAT_INTR_NOTACT)
2609                         return;
2610         }
2611
2612         /*
2613          * NOTE:
2614          * Interrupt will have to be disabled if tagged status
2615          * is used, else interrupt will always be asserted on
2616          * certain chips (at least on BCM5750 AX/BX).
2617          */
2618         bnx_writembx(sc, BGE_MBX_IRQ0_LO, 1);
2619
2620         bnx_intr(sc);
2621 }
2622
2623 static void
2624 bnx_msi(void *xsc)
2625 {
2626         struct bnx_softc *sc = xsc;
2627
2628         /* Disable interrupt first */
2629         bnx_writembx(sc, BGE_MBX_IRQ0_LO, 1);
2630         bnx_intr(sc);
2631 }
2632
2633 static void
2634 bnx_msi_oneshot(void *xsc)
2635 {
2636         bnx_intr(xsc);
2637 }
2638
2639 static void
2640 bnx_intr(struct bnx_softc *sc)
2641 {
2642         struct ifnet *ifp = &sc->arpcom.ac_if;
2643         struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
2644         uint16_t rx_prod, tx_cons;
2645         uint32_t status;
2646
2647         sc->bnx_status_tag = sblk->bge_status_tag;
2648         /*
2649          * Use a load fence to ensure that status_tag is saved 
2650          * before rx_prod, tx_cons and status.
2651          */
2652         cpu_lfence();
2653
2654         rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
2655         tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
2656         status = sblk->bge_status;
2657
2658         if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) || sc->bnx_link_evt)
2659                 bnx_link_poll(sc);
2660
2661         if (ifp->if_flags & IFF_RUNNING) {
2662                 if (sc->bnx_rx_saved_considx != rx_prod)
2663                         bnx_rxeof(sc, rx_prod, -1);
2664
2665                 if (sc->bnx_tx_saved_considx != tx_cons)
2666                         bnx_txeof(sc, tx_cons);
2667         }
2668
2669         bnx_writembx(sc, BGE_MBX_IRQ0_LO, sc->bnx_status_tag << 24);
2670
2671         if (sc->bnx_coal_chg)
2672                 bnx_coal_change(sc);
2673 }
2674
2675 static void
2676 bnx_tick(void *xsc)
2677 {
2678         struct bnx_softc *sc = xsc;
2679         struct ifnet *ifp = &sc->arpcom.ac_if;
2680
2681         lwkt_serialize_enter(ifp->if_serializer);
2682
2683         KKASSERT(mycpuid == sc->bnx_stat_cpuid);
2684
2685         bnx_stats_update_regs(sc);
2686
2687         if (sc->bnx_flags & BNX_FLAG_TBI) {
2688                 /*
2689                  * Since in TBI mode auto-polling can't be used we should poll
2690                  * link status manually. Here we register pending link event
2691                  * and trigger interrupt.
2692                  */
2693                 sc->bnx_link_evt++;
2694                 BNX_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
2695         } else if (!sc->bnx_link) {
2696                 mii_tick(device_get_softc(sc->bnx_miibus));
2697         }
2698
2699         callout_reset(&sc->bnx_stat_timer, hz, bnx_tick, sc);
2700
2701         lwkt_serialize_exit(ifp->if_serializer);
2702 }
2703
2704 static void
2705 bnx_stats_update_regs(struct bnx_softc *sc)
2706 {
2707         struct ifnet *ifp = &sc->arpcom.ac_if;
2708         struct bge_mac_stats_regs stats;
2709         uint32_t *s;
2710         int i;
2711
2712         s = (uint32_t *)&stats;
2713         for (i = 0; i < sizeof(struct bge_mac_stats_regs); i += 4) {
2714                 *s = CSR_READ_4(sc, BGE_RX_STATS + i);
2715                 s++;
2716         }
2717
2718         IFNET_STAT_SET(ifp, collisions,
2719            (stats.dot3StatsSingleCollisionFrames +
2720            stats.dot3StatsMultipleCollisionFrames +
2721            stats.dot3StatsExcessiveCollisions +
2722            stats.dot3StatsLateCollisions));
2723 }
2724
2725 /*
2726  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
2727  * pointers to descriptors.
2728  */
2729 static int
2730 bnx_encap(struct bnx_softc *sc, struct mbuf **m_head0, uint32_t *txidx,
2731     int *segs_used)
2732 {
2733         struct bge_tx_bd *d = NULL;
2734         uint16_t csum_flags = 0, vlan_tag = 0, mss = 0;
2735         bus_dma_segment_t segs[BNX_NSEG_NEW];
2736         bus_dmamap_t map;
2737         int error, maxsegs, nsegs, idx, i;
2738         struct mbuf *m_head = *m_head0, *m_new;
2739
2740         if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
2741 #ifdef BNX_TSO_DEBUG
2742                 int tso_nsegs;
2743 #endif
2744
2745                 error = bnx_setup_tso(sc, m_head0, &mss, &csum_flags);
2746                 if (error)
2747                         return error;
2748                 m_head = *m_head0;
2749
2750 #ifdef BNX_TSO_DEBUG
2751                 tso_nsegs = (m_head->m_pkthdr.len /
2752                     m_head->m_pkthdr.tso_segsz) - 1;
2753                 if (tso_nsegs > (BNX_TSO_NSTATS - 1))
2754                         tso_nsegs = BNX_TSO_NSTATS - 1;
2755                 else if (tso_nsegs < 0)
2756                         tso_nsegs = 0;
2757                 sc->bnx_tsosegs[tso_nsegs]++;
2758 #endif
2759         } else if (m_head->m_pkthdr.csum_flags & BNX_CSUM_FEATURES) {
2760                 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
2761                         csum_flags |= BGE_TXBDFLAG_IP_CSUM;
2762                 if (m_head->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
2763                         csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
2764                 if (m_head->m_flags & M_LASTFRAG)
2765                         csum_flags |= BGE_TXBDFLAG_IP_FRAG_END;
2766                 else if (m_head->m_flags & M_FRAG)
2767                         csum_flags |= BGE_TXBDFLAG_IP_FRAG;
2768         }
2769         if (m_head->m_flags & M_VLANTAG) {
2770                 csum_flags |= BGE_TXBDFLAG_VLAN_TAG;
2771                 vlan_tag = m_head->m_pkthdr.ether_vlantag;
2772         }
2773
2774         idx = *txidx;
2775         map = sc->bnx_cdata.bnx_tx_dmamap[idx];
2776
2777         maxsegs = (BGE_TX_RING_CNT - sc->bnx_txcnt) - BNX_NSEG_RSVD;
2778         KASSERT(maxsegs >= BNX_NSEG_SPARE,
2779                 ("not enough segments %d", maxsegs));
2780
2781         if (maxsegs > BNX_NSEG_NEW)
2782                 maxsegs = BNX_NSEG_NEW;
2783
2784         /*
2785          * Pad outbound frame to BGE_MIN_FRAMELEN for an unusual reason.
2786          * The bge hardware will pad out Tx runts to BGE_MIN_FRAMELEN,
2787          * but when such padded frames employ the bge IP/TCP checksum
2788          * offload, the hardware checksum assist gives incorrect results
2789          * (possibly from incorporating its own padding into the UDP/TCP
2790          * checksum; who knows).  If we pad such runts with zeros, the
2791          * onboard checksum comes out correct.
2792          */
2793         if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) &&
2794             m_head->m_pkthdr.len < BNX_MIN_FRAMELEN) {
2795                 error = m_devpad(m_head, BNX_MIN_FRAMELEN);
2796                 if (error)
2797                         goto back;
2798         }
2799
2800         if ((sc->bnx_flags & BNX_FLAG_SHORTDMA) && m_head->m_next != NULL) {
2801                 m_new = bnx_defrag_shortdma(m_head);
2802                 if (m_new == NULL) {
2803                         error = ENOBUFS;
2804                         goto back;
2805                 }
2806                 *m_head0 = m_head = m_new;
2807         }
2808         if ((m_head->m_pkthdr.csum_flags & CSUM_TSO) == 0 &&
2809             sc->bnx_force_defrag && m_head->m_next != NULL) {
2810                 /*
2811                  * Forcefully defragment mbuf chain to overcome hardware
2812                  * limitation which only support a single outstanding
2813                  * DMA read operation.  If it fails, keep moving on using
2814                  * the original mbuf chain.
2815                  */
2816                 m_new = m_defrag(m_head, MB_DONTWAIT);
2817                 if (m_new != NULL)
2818                         *m_head0 = m_head = m_new;
2819         }
2820
2821         error = bus_dmamap_load_mbuf_defrag(sc->bnx_cdata.bnx_tx_mtag, map,
2822                         m_head0, segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
2823         if (error)
2824                 goto back;
2825         *segs_used += nsegs;
2826
2827         m_head = *m_head0;
2828         bus_dmamap_sync(sc->bnx_cdata.bnx_tx_mtag, map, BUS_DMASYNC_PREWRITE);
2829
2830         for (i = 0; ; i++) {
2831                 d = &sc->bnx_ldata.bnx_tx_ring[idx];
2832
2833                 d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
2834                 d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
2835                 d->bge_len = segs[i].ds_len;
2836                 d->bge_flags = csum_flags;
2837                 d->bge_vlan_tag = vlan_tag;
2838                 d->bge_mss = mss;
2839
2840                 if (i == nsegs - 1)
2841                         break;
2842                 BNX_INC(idx, BGE_TX_RING_CNT);
2843         }
2844         /* Mark the last segment as end of packet... */
2845         d->bge_flags |= BGE_TXBDFLAG_END;
2846
2847         /*
2848          * Insure that the map for this transmission is placed at
2849          * the array index of the last descriptor in this chain.
2850          */
2851         sc->bnx_cdata.bnx_tx_dmamap[*txidx] = sc->bnx_cdata.bnx_tx_dmamap[idx];
2852         sc->bnx_cdata.bnx_tx_dmamap[idx] = map;
2853         sc->bnx_cdata.bnx_tx_chain[idx] = m_head;
2854         sc->bnx_txcnt += nsegs;
2855
2856         BNX_INC(idx, BGE_TX_RING_CNT);
2857         *txidx = idx;
2858 back:
2859         if (error) {
2860                 m_freem(*m_head0);
2861                 *m_head0 = NULL;
2862         }
2863         return error;
2864 }
2865
2866 /*
2867  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
2868  * to the mbuf data regions directly in the transmit descriptors.
2869  */
2870 static void
2871 bnx_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
2872 {
2873         struct bnx_softc *sc = ifp->if_softc;
2874         struct mbuf *m_head = NULL;
2875         uint32_t prodidx;
2876         int nsegs = 0;
2877
2878         ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
2879
2880         if ((ifp->if_flags & IFF_RUNNING) == 0 || ifq_is_oactive(&ifp->if_snd))
2881                 return;
2882
2883         prodidx = sc->bnx_tx_prodidx;
2884
2885         while (sc->bnx_cdata.bnx_tx_chain[prodidx] == NULL) {
2886                 /*
2887                  * Sanity check: avoid coming within BGE_NSEG_RSVD
2888                  * descriptors of the end of the ring.  Also make
2889                  * sure there are BGE_NSEG_SPARE descriptors for
2890                  * jumbo buffers' or TSO segments' defragmentation.
2891                  */
2892                 if ((BGE_TX_RING_CNT - sc->bnx_txcnt) <
2893                     (BNX_NSEG_RSVD + BNX_NSEG_SPARE)) {
2894                         ifq_set_oactive(&ifp->if_snd);
2895                         break;
2896                 }
2897
2898                 m_head = ifq_dequeue(&ifp->if_snd, NULL);
2899                 if (m_head == NULL)
2900                         break;
2901
2902                 /*
2903                  * Pack the data into the transmit ring. If we
2904                  * don't have room, set the OACTIVE flag and wait
2905                  * for the NIC to drain the ring.
2906                  */
2907                 if (bnx_encap(sc, &m_head, &prodidx, &nsegs)) {
2908                         ifq_set_oactive(&ifp->if_snd);
2909                         IFNET_STAT_INC(ifp, oerrors, 1);
2910                         break;
2911                 }
2912
2913                 if (nsegs >= sc->bnx_tx_wreg) {
2914                         /* Transmit */
2915                         bnx_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
2916                         nsegs = 0;
2917                 }
2918
2919                 ETHER_BPF_MTAP(ifp, m_head);
2920
2921                 /*
2922                  * Set a timeout in case the chip goes out to lunch.
2923                  */
2924                 ifp->if_timer = 5;
2925         }
2926
2927         if (nsegs > 0) {
2928                 /* Transmit */
2929                 bnx_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
2930         }
2931         sc->bnx_tx_prodidx = prodidx;
2932 }
2933
2934 static void
2935 bnx_init(void *xsc)
2936 {
2937         struct bnx_softc *sc = xsc;
2938         struct ifnet *ifp = &sc->arpcom.ac_if;
2939         uint16_t *m;
2940         uint32_t mode;
2941
2942         ASSERT_SERIALIZED(ifp->if_serializer);
2943
2944         /* Cancel pending I/O and flush buffers. */
2945         bnx_stop(sc);
2946         bnx_reset(sc);
2947         bnx_chipinit(sc);
2948
2949         /*
2950          * Init the various state machines, ring
2951          * control blocks and firmware.
2952          */
2953         if (bnx_blockinit(sc)) {
2954                 if_printf(ifp, "initialization failure\n");
2955                 bnx_stop(sc);
2956                 return;
2957         }
2958
2959         /* Specify MTU. */
2960         CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
2961             ETHER_HDR_LEN + ETHER_CRC_LEN + EVL_ENCAPLEN);
2962
2963         /* Load our MAC address. */
2964         m = (uint16_t *)&sc->arpcom.ac_enaddr[0];
2965         CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
2966         CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
2967
2968         /* Enable or disable promiscuous mode as needed. */
2969         bnx_setpromisc(sc);
2970
2971         /* Program multicast filter. */
2972         bnx_setmulti(sc);
2973
2974         /* Init RX ring. */
2975         if (bnx_init_rx_ring_std(sc)) {
2976                 if_printf(ifp, "RX ring initialization failed\n");
2977                 bnx_stop(sc);
2978                 return;
2979         }
2980
2981         /* Init jumbo RX ring. */
2982         if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN)) {
2983                 if (bnx_init_rx_ring_jumbo(sc)) {
2984                         if_printf(ifp, "Jumbo RX ring initialization failed\n");
2985                         bnx_stop(sc);
2986                         return;
2987                 }
2988         }
2989
2990         /* Init our RX return ring index */
2991         sc->bnx_rx_saved_considx = 0;
2992
2993         /* Init TX ring. */
2994         bnx_init_tx_ring(sc);
2995
2996         /* Enable TX MAC state machine lockup fix. */
2997         mode = CSR_READ_4(sc, BGE_TX_MODE);
2998         mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
2999         if (sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
3000             sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
3001                 mode &= ~(BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
3002                 mode |= CSR_READ_4(sc, BGE_TX_MODE) &
3003                     (BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
3004         }
3005         /* Turn on transmitter */
3006         CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
3007
3008         /* Turn on receiver */
3009         BNX_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3010
3011         /*
3012          * Set the number of good frames to receive after RX MBUF
3013          * Low Watermark has been reached.  After the RX MAC receives
3014          * this number of frames, it will drop subsequent incoming
3015          * frames until the MBUF High Watermark is reached.
3016          */
3017         if (BNX_IS_57765_FAMILY(sc))
3018                 CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 1);
3019         else
3020                 CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3021
3022         if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI) {
3023                 if (bootverbose) {
3024                         if_printf(ifp, "MSI_MODE: %#x\n",
3025                             CSR_READ_4(sc, BGE_MSI_MODE));
3026                 }
3027         }
3028
3029         /* Tell firmware we're alive. */
3030         BNX_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3031
3032         /* Enable host interrupts if polling(4) is not enabled. */
3033         PCI_SETBIT(sc->bnx_dev, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA, 4);
3034 #ifdef IFPOLL_ENABLE
3035         if (ifp->if_flags & IFF_NPOLLING)
3036                 bnx_disable_intr(sc);
3037         else
3038 #endif
3039         bnx_enable_intr(sc);
3040
3041         bnx_ifmedia_upd(ifp);
3042
3043         ifp->if_flags |= IFF_RUNNING;
3044         ifq_clr_oactive(&ifp->if_snd);
3045
3046         callout_reset_bycpu(&sc->bnx_stat_timer, hz, bnx_tick, sc,
3047             sc->bnx_stat_cpuid);
3048 }
3049
3050 /*
3051  * Set media options.
3052  */
3053 static int
3054 bnx_ifmedia_upd(struct ifnet *ifp)
3055 {
3056         struct bnx_softc *sc = ifp->if_softc;
3057
3058         /* If this is a 1000baseX NIC, enable the TBI port. */
3059         if (sc->bnx_flags & BNX_FLAG_TBI) {
3060                 struct ifmedia *ifm = &sc->bnx_ifmedia;
3061
3062                 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3063                         return(EINVAL);
3064
3065                 switch(IFM_SUBTYPE(ifm->ifm_media)) {
3066                 case IFM_AUTO:
3067                         break;
3068
3069                 case IFM_1000_SX:
3070                         if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
3071                                 BNX_CLRBIT(sc, BGE_MAC_MODE,
3072                                     BGE_MACMODE_HALF_DUPLEX);
3073                         } else {
3074                                 BNX_SETBIT(sc, BGE_MAC_MODE,
3075                                     BGE_MACMODE_HALF_DUPLEX);
3076                         }
3077                         break;
3078                 default:
3079                         return(EINVAL);
3080                 }
3081         } else {
3082                 struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3083
3084                 sc->bnx_link_evt++;
3085                 sc->bnx_link = 0;
3086                 if (mii->mii_instance) {
3087                         struct mii_softc *miisc;
3088
3089                         LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3090                                 mii_phy_reset(miisc);
3091                 }
3092                 mii_mediachg(mii);
3093
3094                 /*
3095                  * Force an interrupt so that we will call bnx_link_upd
3096                  * if needed and clear any pending link state attention.
3097                  * Without this we are not getting any further interrupts
3098                  * for link state changes and thus will not UP the link and
3099                  * not be able to send in bnx_start.  The only way to get
3100                  * things working was to receive a packet and get an RX
3101                  * intr.
3102                  *
3103                  * bnx_tick should help for fiber cards and we might not
3104                  * need to do this here if BNX_FLAG_TBI is set but as
3105                  * we poll for fiber anyway it should not harm.
3106                  */
3107                 BNX_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3108         }
3109         return(0);
3110 }
3111
3112 /*
3113  * Report current media status.
3114  */
3115 static void
3116 bnx_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3117 {
3118         struct bnx_softc *sc = ifp->if_softc;
3119
3120         if (sc->bnx_flags & BNX_FLAG_TBI) {
3121                 ifmr->ifm_status = IFM_AVALID;
3122                 ifmr->ifm_active = IFM_ETHER;
3123                 if (CSR_READ_4(sc, BGE_MAC_STS) &
3124                     BGE_MACSTAT_TBI_PCS_SYNCHED) {
3125                         ifmr->ifm_status |= IFM_ACTIVE;
3126                 } else {
3127                         ifmr->ifm_active |= IFM_NONE;
3128                         return;
3129                 }
3130
3131                 ifmr->ifm_active |= IFM_1000_SX;
3132                 if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
3133                         ifmr->ifm_active |= IFM_HDX;    
3134                 else
3135                         ifmr->ifm_active |= IFM_FDX;
3136         } else {
3137                 struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3138
3139                 mii_pollstat(mii);
3140                 ifmr->ifm_active = mii->mii_media_active;
3141                 ifmr->ifm_status = mii->mii_media_status;
3142         }
3143 }
3144
3145 static int
3146 bnx_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
3147 {
3148         struct bnx_softc *sc = ifp->if_softc;
3149         struct ifreq *ifr = (struct ifreq *)data;
3150         int mask, error = 0;
3151
3152         ASSERT_SERIALIZED(ifp->if_serializer);
3153
3154         switch (command) {
3155         case SIOCSIFMTU:
3156                 if ((!BNX_IS_JUMBO_CAPABLE(sc) && ifr->ifr_mtu > ETHERMTU) ||
3157                     (BNX_IS_JUMBO_CAPABLE(sc) &&
3158                      ifr->ifr_mtu > BNX_JUMBO_MTU)) {
3159                         error = EINVAL;
3160                 } else if (ifp->if_mtu != ifr->ifr_mtu) {
3161                         ifp->if_mtu = ifr->ifr_mtu;
3162                         if (ifp->if_flags & IFF_RUNNING)
3163                                 bnx_init(sc);
3164                 }
3165                 break;
3166         case SIOCSIFFLAGS:
3167                 if (ifp->if_flags & IFF_UP) {
3168                         if (ifp->if_flags & IFF_RUNNING) {
3169                                 mask = ifp->if_flags ^ sc->bnx_if_flags;
3170
3171                                 /*
3172                                  * If only the state of the PROMISC flag
3173                                  * changed, then just use the 'set promisc
3174                                  * mode' command instead of reinitializing
3175                                  * the entire NIC. Doing a full re-init
3176                                  * means reloading the firmware and waiting
3177                                  * for it to start up, which may take a
3178                                  * second or two.  Similarly for ALLMULTI.
3179                                  */
3180                                 if (mask & IFF_PROMISC)
3181                                         bnx_setpromisc(sc);
3182                                 if (mask & IFF_ALLMULTI)
3183                                         bnx_setmulti(sc);
3184                         } else {
3185                                 bnx_init(sc);
3186                         }
3187                 } else if (ifp->if_flags & IFF_RUNNING) {
3188                         bnx_stop(sc);
3189                 }
3190                 sc->bnx_if_flags = ifp->if_flags;
3191                 break;
3192         case SIOCADDMULTI:
3193         case SIOCDELMULTI:
3194                 if (ifp->if_flags & IFF_RUNNING)
3195                         bnx_setmulti(sc);
3196                 break;
3197         case SIOCSIFMEDIA:
3198         case SIOCGIFMEDIA:
3199                 if (sc->bnx_flags & BNX_FLAG_TBI) {
3200                         error = ifmedia_ioctl(ifp, ifr,
3201                             &sc->bnx_ifmedia, command);
3202                 } else {
3203                         struct mii_data *mii;
3204
3205                         mii = device_get_softc(sc->bnx_miibus);
3206                         error = ifmedia_ioctl(ifp, ifr,
3207                                               &mii->mii_media, command);
3208                 }
3209                 break;
3210         case SIOCSIFCAP:
3211                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3212                 if (mask & IFCAP_HWCSUM) {
3213                         ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
3214                         if (ifp->if_capenable & IFCAP_TXCSUM)
3215                                 ifp->if_hwassist |= BNX_CSUM_FEATURES;
3216                         else
3217                                 ifp->if_hwassist &= ~BNX_CSUM_FEATURES;
3218                 }
3219                 if (mask & IFCAP_TSO) {
3220                         ifp->if_capenable ^= (mask & IFCAP_TSO);
3221                         if (ifp->if_capenable & IFCAP_TSO)
3222                                 ifp->if_hwassist |= CSUM_TSO;
3223                         else
3224                                 ifp->if_hwassist &= ~CSUM_TSO;
3225                 }
3226                 break;
3227         default:
3228                 error = ether_ioctl(ifp, command, data);
3229                 break;
3230         }
3231         return error;
3232 }
3233
3234 static void
3235 bnx_watchdog(struct ifnet *ifp)
3236 {
3237         struct bnx_softc *sc = ifp->if_softc;
3238
3239         if_printf(ifp, "watchdog timeout -- resetting\n");
3240
3241         bnx_init(sc);
3242
3243         IFNET_STAT_INC(ifp, oerrors, 1);
3244
3245         if (!ifq_is_empty(&ifp->if_snd))
3246                 if_devstart(ifp);
3247 }
3248
3249 /*
3250  * Stop the adapter and free any mbufs allocated to the
3251  * RX and TX lists.
3252  */
3253 static void
3254 bnx_stop(struct bnx_softc *sc)
3255 {
3256         struct ifnet *ifp = &sc->arpcom.ac_if;
3257
3258         ASSERT_SERIALIZED(ifp->if_serializer);
3259
3260         callout_stop(&sc->bnx_stat_timer);
3261
3262         /*
3263          * Disable all of the receiver blocks
3264          */
3265         bnx_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3266         bnx_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3267         bnx_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
3268         bnx_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
3269         bnx_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
3270         bnx_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
3271
3272         /*
3273          * Disable all of the transmit blocks
3274          */
3275         bnx_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3276         bnx_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3277         bnx_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3278         bnx_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
3279         bnx_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
3280         bnx_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3281
3282         /*
3283          * Shut down all of the memory managers and related
3284          * state machines.
3285          */
3286         bnx_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
3287         bnx_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
3288         CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
3289         CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
3290
3291         /* Disable host interrupts. */
3292         bnx_disable_intr(sc);
3293
3294         /*
3295          * Tell firmware we're shutting down.
3296          */
3297         BNX_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3298
3299         /* Free the RX lists. */
3300         bnx_free_rx_ring_std(sc);
3301
3302         /* Free jumbo RX list. */
3303         if (BNX_IS_JUMBO_CAPABLE(sc))
3304                 bnx_free_rx_ring_jumbo(sc);
3305
3306         /* Free TX buffers. */
3307         bnx_free_tx_ring(sc);
3308
3309         sc->bnx_status_tag = 0;
3310         sc->bnx_link = 0;
3311         sc->bnx_coal_chg = 0;
3312
3313         sc->bnx_tx_saved_considx = BNX_TXCONS_UNSET;
3314
3315         ifp->if_flags &= ~IFF_RUNNING;
3316         ifq_clr_oactive(&ifp->if_snd);
3317         ifp->if_timer = 0;
3318 }
3319
3320 /*
3321  * Stop all chip I/O so that the kernel's probe routines don't
3322  * get confused by errant DMAs when rebooting.
3323  */
3324 static void
3325 bnx_shutdown(device_t dev)
3326 {
3327         struct bnx_softc *sc = device_get_softc(dev);
3328         struct ifnet *ifp = &sc->arpcom.ac_if;
3329
3330         lwkt_serialize_enter(ifp->if_serializer);
3331         bnx_stop(sc);
3332         bnx_reset(sc);
3333         lwkt_serialize_exit(ifp->if_serializer);
3334 }
3335
3336 static int
3337 bnx_suspend(device_t dev)
3338 {
3339         struct bnx_softc *sc = device_get_softc(dev);
3340         struct ifnet *ifp = &sc->arpcom.ac_if;
3341
3342         lwkt_serialize_enter(ifp->if_serializer);
3343         bnx_stop(sc);
3344         lwkt_serialize_exit(ifp->if_serializer);
3345
3346         return 0;
3347 }
3348
3349 static int
3350 bnx_resume(device_t dev)
3351 {
3352         struct bnx_softc *sc = device_get_softc(dev);
3353         struct ifnet *ifp = &sc->arpcom.ac_if;
3354
3355         lwkt_serialize_enter(ifp->if_serializer);
3356
3357         if (ifp->if_flags & IFF_UP) {
3358                 bnx_init(sc);
3359
3360                 if (!ifq_is_empty(&ifp->if_snd))
3361                         if_devstart(ifp);
3362         }
3363
3364         lwkt_serialize_exit(ifp->if_serializer);
3365
3366         return 0;
3367 }
3368
3369 static void
3370 bnx_setpromisc(struct bnx_softc *sc)
3371 {
3372         struct ifnet *ifp = &sc->arpcom.ac_if;
3373
3374         if (ifp->if_flags & IFF_PROMISC)
3375                 BNX_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3376         else
3377                 BNX_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3378 }
3379
3380 static void
3381 bnx_dma_free(struct bnx_softc *sc)
3382 {
3383         int i;
3384
3385         /* Destroy RX mbuf DMA stuffs. */
3386         if (sc->bnx_cdata.bnx_rx_mtag != NULL) {
3387                 for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
3388                         bus_dmamap_destroy(sc->bnx_cdata.bnx_rx_mtag,
3389                             sc->bnx_cdata.bnx_rx_std_dmamap[i]);
3390                 }
3391                 bus_dmamap_destroy(sc->bnx_cdata.bnx_rx_mtag,
3392                                    sc->bnx_cdata.bnx_rx_tmpmap);
3393                 bus_dma_tag_destroy(sc->bnx_cdata.bnx_rx_mtag);
3394         }
3395
3396         /* Destroy TX mbuf DMA stuffs. */
3397         if (sc->bnx_cdata.bnx_tx_mtag != NULL) {
3398                 for (i = 0; i < BGE_TX_RING_CNT; i++) {
3399                         bus_dmamap_destroy(sc->bnx_cdata.bnx_tx_mtag,
3400                             sc->bnx_cdata.bnx_tx_dmamap[i]);
3401                 }
3402                 bus_dma_tag_destroy(sc->bnx_cdata.bnx_tx_mtag);
3403         }
3404
3405         /* Destroy standard RX ring */
3406         bnx_dma_block_free(sc->bnx_cdata.bnx_rx_std_ring_tag,
3407                            sc->bnx_cdata.bnx_rx_std_ring_map,
3408                            sc->bnx_ldata.bnx_rx_std_ring);
3409
3410         if (BNX_IS_JUMBO_CAPABLE(sc))
3411                 bnx_free_jumbo_mem(sc);
3412
3413         /* Destroy RX return ring */
3414         bnx_dma_block_free(sc->bnx_cdata.bnx_rx_return_ring_tag,
3415                            sc->bnx_cdata.bnx_rx_return_ring_map,
3416                            sc->bnx_ldata.bnx_rx_return_ring);
3417
3418         /* Destroy TX ring */
3419         bnx_dma_block_free(sc->bnx_cdata.bnx_tx_ring_tag,
3420                            sc->bnx_cdata.bnx_tx_ring_map,
3421                            sc->bnx_ldata.bnx_tx_ring);
3422
3423         /* Destroy status block */
3424         bnx_dma_block_free(sc->bnx_cdata.bnx_status_tag,
3425                            sc->bnx_cdata.bnx_status_map,
3426                            sc->bnx_ldata.bnx_status_block);
3427
3428         /* Destroy the parent tag */
3429         if (sc->bnx_cdata.bnx_parent_tag != NULL)
3430                 bus_dma_tag_destroy(sc->bnx_cdata.bnx_parent_tag);
3431 }
3432
3433 static int
3434 bnx_dma_alloc(struct bnx_softc *sc)
3435 {
3436         struct ifnet *ifp = &sc->arpcom.ac_if;
3437         bus_size_t txmaxsz, txmaxsegsz;
3438         int i, error;
3439
3440         /*
3441          * Allocate the parent bus DMA tag appropriate for PCI.
3442          *
3443          * All of the NetExtreme/NetLink controllers have 4GB boundary
3444          * DMA bug.
3445          * Whenever an address crosses a multiple of the 4GB boundary
3446          * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition
3447          * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA
3448          * state machine will lockup and cause the device to hang.
3449          */
3450         error = bus_dma_tag_create(NULL, 1, BGE_DMA_BOUNDARY_4G,
3451                                    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3452                                    NULL, NULL,
3453                                    BUS_SPACE_MAXSIZE_32BIT, 0,
3454                                    BUS_SPACE_MAXSIZE_32BIT,
3455                                    0, &sc->bnx_cdata.bnx_parent_tag);
3456         if (error) {
3457                 if_printf(ifp, "could not allocate parent dma tag\n");
3458                 return error;
3459         }
3460
3461         /*
3462          * Create DMA tag and maps for RX mbufs.
3463          */
3464         error = bus_dma_tag_create(sc->bnx_cdata.bnx_parent_tag, 1, 0,
3465                                    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3466                                    NULL, NULL, MCLBYTES, 1, MCLBYTES,
3467                                    BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
3468                                    &sc->bnx_cdata.bnx_rx_mtag);
3469         if (error) {
3470                 if_printf(ifp, "could not allocate RX mbuf dma tag\n");
3471                 return error;
3472         }
3473
3474         error = bus_dmamap_create(sc->bnx_cdata.bnx_rx_mtag,
3475                                   BUS_DMA_WAITOK, &sc->bnx_cdata.bnx_rx_tmpmap);
3476         if (error) {
3477                 bus_dma_tag_destroy(sc->bnx_cdata.bnx_rx_mtag);
3478                 sc->bnx_cdata.bnx_rx_mtag = NULL;
3479                 return error;
3480         }
3481
3482         for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
3483                 error = bus_dmamap_create(sc->bnx_cdata.bnx_rx_mtag,
3484                                           BUS_DMA_WAITOK,
3485                                           &sc->bnx_cdata.bnx_rx_std_dmamap[i]);
3486                 if (error) {
3487                         int j;
3488
3489                         for (j = 0; j < i; ++j) {
3490                                 bus_dmamap_destroy(sc->bnx_cdata.bnx_rx_mtag,
3491                                         sc->bnx_cdata.bnx_rx_std_dmamap[j]);
3492                         }
3493                         bus_dma_tag_destroy(sc->bnx_cdata.bnx_rx_mtag);
3494                         sc->bnx_cdata.bnx_rx_mtag = NULL;
3495
3496                         if_printf(ifp, "could not create DMA map for RX\n");
3497                         return error;
3498                 }
3499         }
3500
3501         /*
3502          * Create DMA tag and maps for TX mbufs.
3503          */
3504         if (sc->bnx_flags & BNX_FLAG_TSO)
3505                 txmaxsz = IP_MAXPACKET + sizeof(struct ether_vlan_header);
3506         else
3507                 txmaxsz = BNX_JUMBO_FRAMELEN;
3508         if (sc->bnx_asicrev == BGE_ASICREV_BCM57766)
3509                 txmaxsegsz = MCLBYTES;
3510         else
3511                 txmaxsegsz = PAGE_SIZE;
3512         error = bus_dma_tag_create(sc->bnx_cdata.bnx_parent_tag, 1, 0,
3513                                    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3514                                    NULL, NULL,
3515                                    txmaxsz, BNX_NSEG_NEW, txmaxsegsz,
3516                                    BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK |
3517                                    BUS_DMA_ONEBPAGE,
3518                                    &sc->bnx_cdata.bnx_tx_mtag);
3519         if (error) {
3520                 if_printf(ifp, "could not allocate TX mbuf dma tag\n");
3521                 return error;
3522         }
3523
3524         for (i = 0; i < BGE_TX_RING_CNT; i++) {
3525                 error = bus_dmamap_create(sc->bnx_cdata.bnx_tx_mtag,
3526                                           BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
3527                                           &sc->bnx_cdata.bnx_tx_dmamap[i]);
3528                 if (error) {
3529                         int j;
3530
3531                         for (j = 0; j < i; ++j) {
3532                                 bus_dmamap_destroy(sc->bnx_cdata.bnx_tx_mtag,
3533                                         sc->bnx_cdata.bnx_tx_dmamap[j]);
3534                         }
3535                         bus_dma_tag_destroy(sc->bnx_cdata.bnx_tx_mtag);
3536                         sc->bnx_cdata.bnx_tx_mtag = NULL;
3537
3538                         if_printf(ifp, "could not create DMA map for TX\n");
3539                         return error;
3540                 }
3541         }
3542
3543         /*
3544          * Create DMA stuffs for standard RX ring.
3545          */
3546         error = bnx_dma_block_alloc(sc, BGE_STD_RX_RING_SZ,
3547                                     &sc->bnx_cdata.bnx_rx_std_ring_tag,
3548                                     &sc->bnx_cdata.bnx_rx_std_ring_map,
3549                                     (void *)&sc->bnx_ldata.bnx_rx_std_ring,
3550                                     &sc->bnx_ldata.bnx_rx_std_ring_paddr);
3551         if (error) {
3552                 if_printf(ifp, "could not create std RX ring\n");
3553                 return error;
3554         }
3555
3556         /*
3557          * Create jumbo buffer pool.
3558          */
3559         if (BNX_IS_JUMBO_CAPABLE(sc)) {
3560                 error = bnx_alloc_jumbo_mem(sc);
3561                 if (error) {
3562                         if_printf(ifp, "could not create jumbo buffer pool\n");
3563                         return error;
3564                 }
3565         }
3566
3567         /*
3568          * Create DMA stuffs for RX return ring.
3569          */
3570         error = bnx_dma_block_alloc(sc,
3571             BGE_RX_RTN_RING_SZ(sc->bnx_return_ring_cnt),
3572             &sc->bnx_cdata.bnx_rx_return_ring_tag,
3573             &sc->bnx_cdata.bnx_rx_return_ring_map,
3574             (void *)&sc->bnx_ldata.bnx_rx_return_ring,
3575             &sc->bnx_ldata.bnx_rx_return_ring_paddr);
3576         if (error) {
3577                 if_printf(ifp, "could not create RX ret ring\n");
3578                 return error;
3579         }
3580
3581         /*
3582          * Create DMA stuffs for TX ring.
3583          */
3584         error = bnx_dma_block_alloc(sc, BGE_TX_RING_SZ,
3585                                     &sc->bnx_cdata.bnx_tx_ring_tag,
3586                                     &sc->bnx_cdata.bnx_tx_ring_map,
3587                                     (void *)&sc->bnx_ldata.bnx_tx_ring,
3588                                     &sc->bnx_ldata.bnx_tx_ring_paddr);
3589         if (error) {
3590                 if_printf(ifp, "could not create TX ring\n");
3591                 return error;
3592         }
3593
3594         /*
3595          * Create DMA stuffs for status block.
3596          */
3597         error = bnx_dma_block_alloc(sc, BGE_STATUS_BLK_SZ,
3598                                     &sc->bnx_cdata.bnx_status_tag,
3599                                     &sc->bnx_cdata.bnx_status_map,
3600                                     (void *)&sc->bnx_ldata.bnx_status_block,
3601                                     &sc->bnx_ldata.bnx_status_block_paddr);
3602         if (error) {
3603                 if_printf(ifp, "could not create status block\n");
3604                 return error;
3605         }
3606
3607         return 0;
3608 }
3609
3610 static int
3611 bnx_dma_block_alloc(struct bnx_softc *sc, bus_size_t size, bus_dma_tag_t *tag,
3612                     bus_dmamap_t *map, void **addr, bus_addr_t *paddr)
3613 {
3614         bus_dmamem_t dmem;
3615         int error;
3616
3617         error = bus_dmamem_coherent(sc->bnx_cdata.bnx_parent_tag, PAGE_SIZE, 0,
3618                                     BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3619                                     size, BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
3620         if (error)
3621                 return error;
3622
3623         *tag = dmem.dmem_tag;
3624         *map = dmem.dmem_map;
3625         *addr = dmem.dmem_addr;
3626         *paddr = dmem.dmem_busaddr;
3627
3628         return 0;
3629 }
3630
3631 static void
3632 bnx_dma_block_free(bus_dma_tag_t tag, bus_dmamap_t map, void *addr)
3633 {
3634         if (tag != NULL) {
3635                 bus_dmamap_unload(tag, map);
3636                 bus_dmamem_free(tag, addr, map);
3637                 bus_dma_tag_destroy(tag);
3638         }
3639 }
3640
3641 static void
3642 bnx_tbi_link_upd(struct bnx_softc *sc, uint32_t status)
3643 {
3644         struct ifnet *ifp = &sc->arpcom.ac_if;
3645
3646 #define PCS_ENCODE_ERR  (BGE_MACSTAT_PORT_DECODE_ERROR|BGE_MACSTAT_MI_COMPLETE)
3647
3648         /*
3649          * Sometimes PCS encoding errors are detected in
3650          * TBI mode (on fiber NICs), and for some reason
3651          * the chip will signal them as link changes.
3652          * If we get a link change event, but the 'PCS
3653          * encoding error' bit in the MAC status register
3654          * is set, don't bother doing a link check.
3655          * This avoids spurious "gigabit link up" messages
3656          * that sometimes appear on fiber NICs during
3657          * periods of heavy traffic.
3658          */
3659         if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
3660                 if (!sc->bnx_link) {
3661                         sc->bnx_link++;
3662                         if (sc->bnx_asicrev == BGE_ASICREV_BCM5704) {
3663                                 BNX_CLRBIT(sc, BGE_MAC_MODE,
3664                                     BGE_MACMODE_TBI_SEND_CFGS);
3665                         }
3666                         CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
3667
3668                         if (bootverbose)
3669                                 if_printf(ifp, "link UP\n");
3670
3671                         ifp->if_link_state = LINK_STATE_UP;
3672                         if_link_state_change(ifp);
3673                 }
3674         } else if ((status & PCS_ENCODE_ERR) != PCS_ENCODE_ERR) {
3675                 if (sc->bnx_link) {
3676                         sc->bnx_link = 0;
3677
3678                         if (bootverbose)
3679                                 if_printf(ifp, "link DOWN\n");
3680
3681                         ifp->if_link_state = LINK_STATE_DOWN;
3682                         if_link_state_change(ifp);
3683                 }
3684         }
3685
3686 #undef PCS_ENCODE_ERR
3687
3688         /* Clear the attention. */
3689         CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3690             BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3691             BGE_MACSTAT_LINK_CHANGED);
3692 }
3693
3694 static void
3695 bnx_copper_link_upd(struct bnx_softc *sc, uint32_t status __unused)
3696 {
3697         struct ifnet *ifp = &sc->arpcom.ac_if;
3698         struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3699
3700         mii_pollstat(mii);
3701         bnx_miibus_statchg(sc->bnx_dev);
3702
3703         if (bootverbose) {
3704                 if (sc->bnx_link)
3705                         if_printf(ifp, "link UP\n");
3706                 else
3707                         if_printf(ifp, "link DOWN\n");
3708         }
3709
3710         /* Clear the attention. */
3711         CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3712             BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3713             BGE_MACSTAT_LINK_CHANGED);
3714 }
3715
3716 static void
3717 bnx_autopoll_link_upd(struct bnx_softc *sc, uint32_t status __unused)
3718 {
3719         struct ifnet *ifp = &sc->arpcom.ac_if;
3720         struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3721
3722         mii_pollstat(mii);
3723
3724         if (!sc->bnx_link &&
3725             (mii->mii_media_status & IFM_ACTIVE) &&
3726             IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
3727                 sc->bnx_link++;
3728                 if (bootverbose)
3729                         if_printf(ifp, "link UP\n");
3730         } else if (sc->bnx_link &&
3731             (!(mii->mii_media_status & IFM_ACTIVE) ||
3732             IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
3733                 sc->bnx_link = 0;
3734                 if (bootverbose)
3735                         if_printf(ifp, "link DOWN\n");
3736         }
3737
3738         /* Clear the attention. */
3739         CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3740             BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3741             BGE_MACSTAT_LINK_CHANGED);
3742 }
3743
3744 static int
3745 bnx_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS)
3746 {
3747         struct bnx_softc *sc = arg1;
3748
3749         return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3750             &sc->bnx_rx_coal_ticks,
3751             BNX_RX_COAL_TICKS_MIN, BNX_RX_COAL_TICKS_MAX,
3752             BNX_RX_COAL_TICKS_CHG);
3753 }
3754
3755 static int
3756 bnx_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS)
3757 {
3758         struct bnx_softc *sc = arg1;
3759
3760         return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3761             &sc->bnx_tx_coal_ticks,
3762             BNX_TX_COAL_TICKS_MIN, BNX_TX_COAL_TICKS_MAX,
3763             BNX_TX_COAL_TICKS_CHG);
3764 }
3765
3766 static int
3767 bnx_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS)
3768 {
3769         struct bnx_softc *sc = arg1;
3770
3771         return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3772             &sc->bnx_rx_coal_bds,
3773             BNX_RX_COAL_BDS_MIN, BNX_RX_COAL_BDS_MAX,
3774             BNX_RX_COAL_BDS_CHG);
3775 }
3776
3777 static int
3778 bnx_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS)
3779 {
3780         struct bnx_softc *sc = arg1;
3781
3782         return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3783             &sc->bnx_tx_coal_bds,
3784             BNX_TX_COAL_BDS_MIN, BNX_TX_COAL_BDS_MAX,
3785             BNX_TX_COAL_BDS_CHG);
3786 }
3787
3788 static int
3789 bnx_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS)
3790 {
3791         struct bnx_softc *sc = arg1;
3792
3793         return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3794             &sc->bnx_rx_coal_bds_int,
3795             BNX_RX_COAL_BDS_MIN, BNX_RX_COAL_BDS_MAX,
3796             BNX_RX_COAL_BDS_INT_CHG);
3797 }
3798
3799 static int
3800 bnx_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS)
3801 {
3802         struct bnx_softc *sc = arg1;
3803
3804         return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3805             &sc->bnx_tx_coal_bds_int,
3806             BNX_TX_COAL_BDS_MIN, BNX_TX_COAL_BDS_MAX,
3807             BNX_TX_COAL_BDS_INT_CHG);
3808 }
3809
3810 static int
3811 bnx_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *coal,
3812     int coal_min, int coal_max, uint32_t coal_chg_mask)
3813 {
3814         struct bnx_softc *sc = arg1;
3815         struct ifnet *ifp = &sc->arpcom.ac_if;
3816         int error = 0, v;
3817
3818         lwkt_serialize_enter(ifp->if_serializer);
3819
3820         v = *coal;
3821         error = sysctl_handle_int(oidp, &v, 0, req);
3822         if (!error && req->newptr != NULL) {
3823                 if (v < coal_min || v > coal_max) {
3824                         error = EINVAL;
3825                 } else {
3826                         *coal = v;
3827                         sc->bnx_coal_chg |= coal_chg_mask;
3828                 }
3829         }
3830
3831         lwkt_serialize_exit(ifp->if_serializer);
3832         return error;
3833 }
3834
3835 static void
3836 bnx_coal_change(struct bnx_softc *sc)
3837 {
3838         struct ifnet *ifp = &sc->arpcom.ac_if;
3839
3840         ASSERT_SERIALIZED(ifp->if_serializer);
3841
3842         if (sc->bnx_coal_chg & BNX_RX_COAL_TICKS_CHG) {
3843                 CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS,
3844                             sc->bnx_rx_coal_ticks);
3845                 DELAY(10);
3846                 CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
3847
3848                 if (bootverbose) {
3849                         if_printf(ifp, "rx_coal_ticks -> %u\n",
3850                                   sc->bnx_rx_coal_ticks);
3851                 }
3852         }
3853
3854         if (sc->bnx_coal_chg & BNX_TX_COAL_TICKS_CHG) {
3855                 CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS,
3856                             sc->bnx_tx_coal_ticks);
3857                 DELAY(10);
3858                 CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS);
3859
3860                 if (bootverbose) {
3861                         if_printf(ifp, "tx_coal_ticks -> %u\n",
3862                                   sc->bnx_tx_coal_ticks);
3863                 }
3864         }
3865
3866         if (sc->bnx_coal_chg & BNX_RX_COAL_BDS_CHG) {
3867                 CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS,
3868                             sc->bnx_rx_coal_bds);
3869                 DELAY(10);
3870                 CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
3871
3872                 if (bootverbose) {
3873                         if_printf(ifp, "rx_coal_bds -> %u\n",
3874                                   sc->bnx_rx_coal_bds);
3875                 }
3876         }
3877
3878         if (sc->bnx_coal_chg & BNX_TX_COAL_BDS_CHG) {
3879                 CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS,
3880                             sc->bnx_tx_coal_bds);
3881                 DELAY(10);
3882                 CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS);
3883
3884                 if (bootverbose) {
3885                         if_printf(ifp, "tx_coal_bds -> %u\n",
3886                                   sc->bnx_tx_coal_bds);
3887                 }
3888         }
3889
3890         if (sc->bnx_coal_chg & BNX_RX_COAL_BDS_INT_CHG) {
3891                 CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT,
3892                     sc->bnx_rx_coal_bds_int);
3893                 DELAY(10);
3894                 CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT);
3895
3896                 if (bootverbose) {
3897                         if_printf(ifp, "rx_coal_bds_int -> %u\n",
3898                             sc->bnx_rx_coal_bds_int);
3899                 }
3900         }
3901
3902         if (sc->bnx_coal_chg & BNX_TX_COAL_BDS_INT_CHG) {
3903                 CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT,
3904                     sc->bnx_tx_coal_bds_int);
3905                 DELAY(10);
3906                 CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT);
3907
3908                 if (bootverbose) {
3909                         if_printf(ifp, "tx_coal_bds_int -> %u\n",
3910                             sc->bnx_tx_coal_bds_int);
3911                 }
3912         }
3913
3914         sc->bnx_coal_chg = 0;
3915 }
3916
3917 static void
3918 bnx_intr_check(void *xsc)
3919 {
3920         struct bnx_softc *sc = xsc;
3921         struct ifnet *ifp = &sc->arpcom.ac_if;
3922         struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
3923
3924         lwkt_serialize_enter(ifp->if_serializer);
3925
3926         KKASSERT(mycpuid == sc->bnx_intr_cpuid);
3927
3928         if ((ifp->if_flags & (IFF_RUNNING | IFF_NPOLLING)) != IFF_RUNNING) {
3929                 lwkt_serialize_exit(ifp->if_serializer);
3930                 return;
3931         }
3932
3933         if (sblk->bge_idx[0].bge_rx_prod_idx != sc->bnx_rx_saved_considx ||
3934             sblk->bge_idx[0].bge_tx_cons_idx != sc->bnx_tx_saved_considx) {
3935                 if (sc->bnx_rx_check_considx == sc->bnx_rx_saved_considx &&
3936                     sc->bnx_tx_check_considx == sc->bnx_tx_saved_considx) {
3937                         if (!sc->bnx_intr_maylose) {
3938                                 sc->bnx_intr_maylose = TRUE;
3939                                 goto done;
3940                         }
3941                         if (bootverbose)
3942                                 if_printf(ifp, "lost interrupt\n");
3943                         bnx_msi(sc);
3944                 }
3945         }
3946         sc->bnx_intr_maylose = FALSE;
3947         sc->bnx_rx_check_considx = sc->bnx_rx_saved_considx;
3948         sc->bnx_tx_check_considx = sc->bnx_tx_saved_considx;
3949
3950 done:
3951         callout_reset(&sc->bnx_intr_timer, BNX_INTR_CKINTVL,
3952             bnx_intr_check, sc);
3953         lwkt_serialize_exit(ifp->if_serializer);
3954 }
3955
3956 static void
3957 bnx_enable_intr(struct bnx_softc *sc)
3958 {
3959         struct ifnet *ifp = &sc->arpcom.ac_if;
3960
3961         lwkt_serialize_handler_enable(ifp->if_serializer);
3962
3963         /*
3964          * Enable interrupt.
3965          */
3966         bnx_writembx(sc, BGE_MBX_IRQ0_LO, sc->bnx_status_tag << 24);
3967         if (sc->bnx_flags & BNX_FLAG_ONESHOT_MSI) {
3968                 /* XXX Linux driver */
3969                 bnx_writembx(sc, BGE_MBX_IRQ0_LO, sc->bnx_status_tag << 24);
3970         }
3971
3972         /*
3973          * Unmask the interrupt when we stop polling.
3974          */
3975         PCI_CLRBIT(sc->bnx_dev, BGE_PCI_MISC_CTL,
3976             BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
3977
3978         /*
3979          * Trigger another interrupt, since above writing
3980          * to interrupt mailbox0 may acknowledge pending
3981          * interrupt.
3982          */
3983         BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3984
3985         if (sc->bnx_flags & BNX_FLAG_STATUSTAG_BUG) {
3986                 sc->bnx_intr_maylose = FALSE;
3987                 sc->bnx_rx_check_considx = 0;
3988                 sc->bnx_tx_check_considx = 0;
3989
3990                 if (bootverbose)
3991                         if_printf(ifp, "status tag bug workaround\n");
3992
3993                 /* 10ms check interval */
3994                 callout_reset_bycpu(&sc->bnx_intr_timer, BNX_INTR_CKINTVL,
3995                     bnx_intr_check, sc, sc->bnx_intr_cpuid);
3996         }
3997 }
3998
3999 static void
4000 bnx_disable_intr(struct bnx_softc *sc)
4001 {
4002         struct ifnet *ifp = &sc->arpcom.ac_if;
4003
4004         /*
4005          * Mask the interrupt when we start polling.
4006          */
4007         PCI_SETBIT(sc->bnx_dev, BGE_PCI_MISC_CTL,
4008             BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4009
4010         /*
4011          * Acknowledge possible asserted interrupt.
4012          */
4013         bnx_writembx(sc, BGE_MBX_IRQ0_LO, 1);
4014
4015         callout_stop(&sc->bnx_intr_timer);
4016         sc->bnx_intr_maylose = FALSE;
4017         sc->bnx_rx_check_considx = 0;
4018         sc->bnx_tx_check_considx = 0;
4019
4020         sc->bnx_npoll.ifpc_stcount = 0;
4021
4022         lwkt_serialize_handler_disable(ifp->if_serializer);
4023 }
4024
4025 static int
4026 bnx_get_eaddr_mem(struct bnx_softc *sc, uint8_t ether_addr[])
4027 {
4028         uint32_t mac_addr;
4029         int ret = 1;
4030
4031         mac_addr = bnx_readmem_ind(sc, 0x0c14);
4032         if ((mac_addr >> 16) == 0x484b) {
4033                 ether_addr[0] = (uint8_t)(mac_addr >> 8);
4034                 ether_addr[1] = (uint8_t)mac_addr;
4035                 mac_addr = bnx_readmem_ind(sc, 0x0c18);
4036                 ether_addr[2] = (uint8_t)(mac_addr >> 24);
4037                 ether_addr[3] = (uint8_t)(mac_addr >> 16);
4038                 ether_addr[4] = (uint8_t)(mac_addr >> 8);
4039                 ether_addr[5] = (uint8_t)mac_addr;
4040                 ret = 0;
4041         }
4042         return ret;
4043 }
4044
4045 static int
4046 bnx_get_eaddr_nvram(struct bnx_softc *sc, uint8_t ether_addr[])
4047 {
4048         int mac_offset = BGE_EE_MAC_OFFSET;
4049
4050         if (BNX_IS_5717_PLUS(sc)) {
4051                 int f;
4052
4053                 f = pci_get_function(sc->bnx_dev);
4054                 if (f & 1)
4055                         mac_offset = BGE_EE_MAC_OFFSET_5717;
4056                 if (f > 1)
4057                         mac_offset += BGE_EE_MAC_OFFSET_5717_OFF;
4058         }
4059
4060         return bnx_read_nvram(sc, ether_addr, mac_offset + 2, ETHER_ADDR_LEN);
4061 }
4062
4063 static int
4064 bnx_get_eaddr_eeprom(struct bnx_softc *sc, uint8_t ether_addr[])
4065 {
4066         if (sc->bnx_flags & BNX_FLAG_NO_EEPROM)
4067                 return 1;
4068
4069         return bnx_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4070                                ETHER_ADDR_LEN);
4071 }
4072
4073 static int
4074 bnx_get_eaddr(struct bnx_softc *sc, uint8_t eaddr[])
4075 {
4076         static const bnx_eaddr_fcn_t bnx_eaddr_funcs[] = {
4077                 /* NOTE: Order is critical */
4078                 bnx_get_eaddr_mem,
4079                 bnx_get_eaddr_nvram,
4080                 bnx_get_eaddr_eeprom,
4081                 NULL
4082         };
4083         const bnx_eaddr_fcn_t *func;
4084
4085         for (func = bnx_eaddr_funcs; *func != NULL; ++func) {
4086                 if ((*func)(sc, eaddr) == 0)
4087                         break;
4088         }
4089         return (*func == NULL ? ENXIO : 0);
4090 }
4091
4092 /*
4093  * NOTE: 'm' is not freed upon failure
4094  */
4095 struct mbuf *
4096 bnx_defrag_shortdma(struct mbuf *m)
4097 {
4098         struct mbuf *n;
4099         int found;
4100
4101         /*
4102          * If device receive two back-to-back send BDs with less than
4103          * or equal to 8 total bytes then the device may hang.  The two
4104          * back-to-back send BDs must in the same frame for this failure
4105          * to occur.  Scan mbuf chains and see whether two back-to-back
4106          * send BDs are there.  If this is the case, allocate new mbuf
4107          * and copy the frame to workaround the silicon bug.
4108          */
4109         for (n = m, found = 0; n != NULL; n = n->m_next) {
4110                 if (n->m_len < 8) {
4111                         found++;
4112                         if (found > 1)
4113                                 break;
4114                         continue;
4115                 }
4116                 found = 0;
4117         }
4118
4119         if (found > 1)
4120                 n = m_defrag(m, MB_DONTWAIT);
4121         else
4122                 n = m;
4123         return n;
4124 }
4125
4126 static void
4127 bnx_stop_block(struct bnx_softc *sc, bus_size_t reg, uint32_t bit)
4128 {
4129         int i;
4130
4131         BNX_CLRBIT(sc, reg, bit);
4132         for (i = 0; i < BNX_TIMEOUT; i++) {
4133                 if ((CSR_READ_4(sc, reg) & bit) == 0)
4134                         return;
4135                 DELAY(100);
4136         }
4137 }
4138
4139 static void
4140 bnx_link_poll(struct bnx_softc *sc)
4141 {
4142         uint32_t status;
4143
4144         status = CSR_READ_4(sc, BGE_MAC_STS);
4145         if ((status & sc->bnx_link_chg) || sc->bnx_link_evt) {
4146                 sc->bnx_link_evt = 0;
4147                 sc->bnx_link_upd(sc, status);
4148         }
4149 }
4150
4151 static void
4152 bnx_enable_msi(struct bnx_softc *sc)
4153 {
4154         uint32_t msi_mode;
4155
4156         msi_mode = CSR_READ_4(sc, BGE_MSI_MODE);
4157         msi_mode |= BGE_MSIMODE_ENABLE;
4158         if (sc->bnx_flags & BNX_FLAG_ONESHOT_MSI) {
4159                 /*
4160                  * NOTE:
4161                  * 5718-PG105-R says that "one shot" mode
4162                  * does not work if MSI is used, however,
4163                  * it obviously works.
4164                  */
4165                 msi_mode &= ~BGE_MSIMODE_ONESHOT_DISABLE;
4166         }
4167         CSR_WRITE_4(sc, BGE_MSI_MODE, msi_mode);
4168 }
4169
4170 static uint32_t
4171 bnx_dma_swap_options(struct bnx_softc *sc)
4172 {
4173         uint32_t dma_options;
4174
4175         dma_options = BGE_MODECTL_WORDSWAP_NONFRAME |
4176             BGE_MODECTL_BYTESWAP_DATA | BGE_MODECTL_WORDSWAP_DATA;
4177 #if BYTE_ORDER == BIG_ENDIAN
4178         dma_options |= BGE_MODECTL_BYTESWAP_NONFRAME;
4179 #endif
4180         if (sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
4181             sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
4182                 dma_options |= BGE_MODECTL_BYTESWAP_B2HRX_DATA |
4183                     BGE_MODECTL_WORDSWAP_B2HRX_DATA | BGE_MODECTL_B2HRX_ENABLE |
4184                     BGE_MODECTL_HTX2B_ENABLE;
4185         }
4186         return dma_options;
4187 }
4188
4189 static int
4190 bnx_setup_tso(struct bnx_softc *sc, struct mbuf **mp,
4191     uint16_t *mss0, uint16_t *flags0)
4192 {
4193         struct mbuf *m;
4194         struct ip *ip;
4195         struct tcphdr *th;
4196         int thoff, iphlen, hoff, hlen;
4197         uint16_t flags, mss;
4198
4199         m = *mp;
4200         KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
4201
4202         hoff = m->m_pkthdr.csum_lhlen;
4203         iphlen = m->m_pkthdr.csum_iphlen;
4204         thoff = m->m_pkthdr.csum_thlen;
4205
4206         KASSERT(hoff > 0, ("invalid ether header len"));
4207         KASSERT(iphlen > 0, ("invalid ip header len"));
4208         KASSERT(thoff > 0, ("invalid tcp header len"));
4209
4210         if (__predict_false(m->m_len < hoff + iphlen + thoff)) {
4211                 m = m_pullup(m, hoff + iphlen + thoff);
4212                 if (m == NULL) {
4213                         *mp = NULL;
4214                         return ENOBUFS;
4215                 }
4216                 *mp = m;
4217         }
4218         ip = mtodoff(m, struct ip *, hoff);
4219         th = mtodoff(m, struct tcphdr *, hoff + iphlen);
4220
4221         mss = m->m_pkthdr.tso_segsz;
4222         flags = BGE_TXBDFLAG_CPU_PRE_DMA | BGE_TXBDFLAG_CPU_POST_DMA;
4223
4224         ip->ip_len = htons(mss + iphlen + thoff);
4225         th->th_sum = 0;
4226
4227         hlen = (iphlen + thoff) >> 2;
4228         mss |= ((hlen & 0x3) << 14);
4229         flags |= ((hlen & 0xf8) << 7) | ((hlen & 0x4) << 2);
4230
4231         *mss0 = mss;
4232         *flags0 = flags;
4233
4234         return 0;
4235 }